University of Kentucky UKnowledge
Theses and Dissertations--Economics Economics
2018
ESSAYS ON THE ROLE OF GOVERNMENT REGULATION AND POLICY IN HEALTH CARE MARKETS
Grayson L. Forlines University of Kentucky, [email protected] Digital Object Identifier: https://doi.org/10.13023/etd.2018.238
Right click to open a feedback form in a new tab to let us know how this document benefits ou.y
Recommended Citation Forlines, Grayson L., "ESSAYS ON THE ROLE OF GOVERNMENT REGULATION AND POLICY IN HEALTH CARE MARKETS" (2018). Theses and Dissertations--Economics. 35. https://uknowledge.uky.edu/economics_etds/35
This Doctoral Dissertation is brought to you for free and open access by the Economics at UKnowledge. It has been accepted for inclusion in Theses and Dissertations--Economics by an authorized administrator of UKnowledge. For more information, please contact [email protected]. STUDENT AGREEMENT:
I represent that my thesis or dissertation and abstract are my original work. Proper attribution has been given to all outside sources. I understand that I am solely responsible for obtaining any needed copyright permissions. I have obtained needed written permission statement(s) from the owner(s) of each third-party copyrighted matter to be included in my work, allowing electronic distribution (if such use is not permitted by the fair use doctrine) which will be submitted to UKnowledge as Additional File.
I hereby grant to The University of Kentucky and its agents the irrevocable, non-exclusive, and royalty-free license to archive and make accessible my work in whole or in part in all forms of media, now or hereafter known. I agree that the document mentioned above may be made available immediately for worldwide access unless an embargo applies.
I retain all other ownership rights to the copyright of my work. I also retain the right to use in future works (such as articles or books) all or part of my work. I understand that I am free to register the copyright to my work.
REVIEW, APPROVAL AND ACCEPTANCE
The document mentioned above has been reviewed and accepted by the student’s advisor, on behalf of the advisory committee, and by the Director of Graduate Studies (DGS), on behalf of the program; we verify that this is the final, approved version of the student’s thesis including all changes required by the advisory committee. The undersigned agree to abide by the statements above.
Grayson L. Forlines, Student
Dr. Frank Scott, Major Professor
Dr. Jenny Minier, Director of Graduate Studies ESSAYS ON THE ROLE OF GOVERNMENT REGULATION AND POLICY IN HEALTH CARE MARKETS
DISSERTATION
A dissertation submitted in partial fulfillment of the requirements for the degree of Doctor of Philosophy in the College of Business and Economics at the University of Kentucky
By Grayson L. Forlines Lexington, Kentucky
Director: Dr. Frank Scott, Professor of Economics Lexington, Kentucky
Copyright© Grayson L. Forlines 2018 ABSTRACT OF DISSERTATION
ESSAYS ON THE ROLE OF GOVERNMENT REGULATION AND POLICY IN HEALTH CARE MARKETS
Understanding how health care markets function is important not only because com- petition has a direct influence on the price and utilization of health care services, but also because the proper functioning, or lack thereof, of health care markets has a very real impact on patients who depend on health care markets and providers for their personal well-being. In this dissertation, I examine the role of government policies and regulation in health care markets, with a focus on the response of health care providers. In Chapter 1, I analyze the impact of Medicare payment rules on hospital ownership of physician practices. Since the mid-2000’s, there has been a rapid increase in hospital ownership of physician practices, however, there is little empirical research which addresses the causes of this re- cent wave of integration. Medicare’s “provider-based” billing policy allows hospital-owned physician practices to charge higher reimbursement rates for services provided compared to a freestanding, independent physician practice, without altering how or where services are provided. This “site-based” differential creates a premium for physicians to integrate with hospitals, and the size of this differential varies with the types of health care services provided. I find that Medicare payment rules have contributed to hospital ownership of physician practices and that the response varies across physician specialties. A 10 percent increase in the relative reimbursement rate paid to integrated physicians leads to a 1.9 percentage point increase in the probability of hospital ownership for Medical Care special- ties, including cardiology, neurology, and dermatology, which explains about one-third of observed integration of these specialties from 2005 through 2015. Magnitudes for Surgical Care specialties are similar, but more sensitive across specifications. There is no significant response for Primary Care physicians. In combination with other empirical literature which finds that integration between physicians and hospitals typically results in higher prices with no impact on costs or quality of care, I cautiously interpret this responsiveness as ev- idence that Medicare’s provider-based billing policy overcompensates integrated physician practices and leads to an inefficiently high level of vertical integration between physician and hospitals.
In Chapter 2, I analyze the effect of anti-fraud enforcement activity on Medicaid spend- ing, with a particular focus on the False Claims Act. The False Claims Act (FCA) is a federal statute which protects the government from making undeserved payments to contractors and suppliers. Individual states have chosen to enact their own versions of the federal FCA, and these statutes have increasingly been used to target health care fraud. FCA statutes commonly include substantial monetary penalties such as “per-violation” monetary fines and tripled damages, as well as a “whistleblower” provision which allows private plaintiffs to initiate a lawsuit and collect a portion of recoveries as a reward. Using variation in state- level FCA legislation, I find state FCAs reduce Medicaid prescription drug spending by 21 percent, while other spending categories - which are less lucrative for FCA lawsuits - are un- responsive. Within the prescription drug category, drugs prone to off-label use show larger declines in response to the whistleblower laws, consistent with FCA lawsuits being used to prosecute pharmaceutical manufacturers for off-label marketing and promotion. Spending and prescription volume for drugs prone to off-label use fall by up to 14 percent. This effect could be driven by pharmaceutical manufacturers’ changes in physician detailing for drugs prone to off-label use and/or physicians’ changes in prescribing behavior.
KEYWORDS: Medicare; Vertical Integration; Hospital ownership of physician practices; Medicaid; False Claims Act; Anti-fraud enforcement
Author’s signature: Grayson L. Forlines
Date: June 26, 2018 ESSAYS ON THE ROLE OF GOVERNMENT REGULATION AND POLICY IN HEALTH CARE MARKETS
By Grayson L. Forlines
Director of Dissertation: Dr. Frank Scott
Director of Graduate Studies: Dr. Jenny Minier
Date: June 26, 2018 For Ashley and Maxwell ACKNOWLEDGMENTS
I am grateful for the guidance, comments, and help provided by my Dissertation Chair, Dr. Frank Scott, in the writing of this dissertation. I thank the members of my disserta- tion committee, Drs. John Garen, Scott Hankins, and Aaron Yelowitz for their comments, contributions, and feedback. This work has also benefited greatly from the collegiality and attention of faculty members and other graduate students in the Department of Economics at the University of Kentucky. I thank the Schnatter Institute for the Study of Free Enter- prise for the fellowship support during the writing of this dissertation, and I am thankful for the generous donation by Mr. John Schnatter which made this possible. A number of people have contributed to my success in the Ph.D program through their support and encouragement. I am especially grateful for the mentoring offered by Michael Brookshire including the opportunity to work at Brookshire, Barrett & Associates. I thank my parents for their support throughout my undergraduate and graduate student career. Finally, I thank my wife, Ashley, for her love, support, and encouragement as I pursued my Ph.D in economics.
iii TABLE OF CONTENTS
Acknowledgments ...... iii
Table of Contents ...... iv
List of Tables ...... vi
List of Figures ...... vii
Chapter 1 Drivers of Physician-Hospital Integration: The Role of Medicare Reim- bursement ...... 1 1.1 Introduction ...... 1 1.2 Background and Literature ...... 4 1.2.1 Integration in the 1990s ...... 4 1.2.2 Integration Today ...... 4 1.2.3 Consequences and Causes of Physician-Hospital Integration . . . . . 6 1.3 Medicare Reimbursement ...... 12 1.3.1 Medicare Payment Systems ...... 14 1.3.2 Federal Anti-fraud and Abuse Regulations ...... 15 1.4 Data ...... 16 1.4.1 Medicare Physician Fee Schedule and Hospital Outpatient Prospec- tive Payment System ...... 16 1.4.2 Medicare Utilization and Payment Data ...... 17 1.4.3 National Ambulatory Medical Care Survey ...... 20 1.5 Model and Methodology ...... 23 1.6 Empirical Results ...... 25 1.7 Robustness ...... 28 1.7.1 Parallel Trends ...... 28 1.7.2 Medicare Patient Share ...... 29 1.7.3 Dynamic Effects ...... 31 1.8 Discussion ...... 33 1.8.1 Geographic Variation in Hospital Ownership ...... 33 1.9 Conclusion ...... 35
Chapter 2 Outsourcing Fraud Enforcement – Whistleblower Laws and Medicaid Expenditure ...... 63 2.1 Introduction ...... 63 2.2 Literature ...... 66 2.2.1 Fraud Committed Against the Government ...... 66 2.2.2 The Social Optimality of Whistleblower Laws ...... 67 2.2.3 Empirical Studies on Health Care Fraud in Public Programs . . . . 67 2.3 The False Claims Act ...... 68 2.3.1 FCA Utilization and Recoveries ...... 69 2.3.2 Recent Amendments ...... 69
iv 2.3.3 FCA Characteristics, Filing Procedure, and the Role of the Whistle- blower ...... 70 2.4 Conceptual Issues: Why Prescription Drugs? ...... 71 2.4.1 FCAs and Different Types of Fraud ...... 71 2.4.2 Policy Confounders ...... 74 2.4.3 State and Federal FCA Overlap ...... 76 2.5 Data and Descriptive Statistics ...... 77 2.6 Empirical Model ...... 83 2.6.1 Difference-in-Difference Model Specification ...... 83 2.6.2 Aggregate Medicaid Expenditure Results ...... 84 2.7 Off-Label Usage: State Drug Utilization Data ...... 87 2.7.1 SDUD ...... 87 2.7.2 Therapeutic Class and Off-Label Use ...... 88 2.7.3 Model ...... 90 2.7.4 Results ...... 91 2.7.5 Robustness ...... 93 2.8 Discussion ...... 96
Appendix A Chapter 1 Appendix ...... 125 A.1 Treatment of Standard Errors in Regression Analyses ...... 125 A.2 Additional Regression Model Specifications ...... 126 A.3 Medicare Patient Share ...... 127 A.4 Relative Prices and Provider Exposure to Provider-Based Billing ...... 128
Appendix B Chapter 2 Appendix ...... 142 B.1 Additional Specifications for Aggregate Medicaid Expenditure Models . . . 142 B.2 Continuous Off-label Rates ...... 145 B.3 Off-label Prone Categorization ...... 145 B.4 Additional Specifications for Models using SDUD ...... 155 B.5 State False Claims Act Legislation ...... 162
References for Chapter 1 ...... 193
References for Chapter 2 ...... 198
Vita ...... 207
v LIST OF TABLES
1.1 Medicare Reimbursement by Specialty ...... 38 1.2 2005 - 2015 NAMCS: Mean Statistics by Practice Ownership ...... 39 1.3 2005 - 2015: Summary Statistics by Specialty ...... 40 1.4 Hospital Ownership of Physician Practices ...... 41 1.5 Hospital Ownership of Physician Practices - Medicare Patient Share Interactions 42 1.6 Hospital Ownership of Physician Practices - Medicare Patient Share Interactions 43 1.7 Hospital Ownership of Physician Practices - Lags and Leads of Relative Price 44 1.8 Hospital Ownership of Physician Practices - Two-Year Lags and Leads of Relative Price ...... 45
2.1 State FCA Legislation Enactment Year ...... 99 2.2 Summary Statistics: 1999 and 2012 ...... 100 2.3 Summary Statistics in 2012 for States With and Without a FCA ...... 101 2.4 Drug Policy Summary Statistics Over Time ...... 102 2.5 Medicaid Spending and Eligibles by Service and Group ...... 103 2.6 MSIS Aggregate Medicaid Expenditure Results ...... 104 2.7 Therapeutic Classes by Number of Prescriptions: SDUD 1999 - 2012 . . . . . 105 2.8 Off-Label “Prone” Classification of Therapeutic Classes ...... 106 2.9 Medical Literature Categorization of Off-label Prone Classes ...... 107 2.10 Continuous Estimated Off-Label Rate Categorization ...... 108 2.11 Dynamic Effects ...... 109
vi LIST OF FIGURES
1.1 Percent of Hospitals Engaged in Affiliations With Physicians ...... 46 1.2 Physicians’ Place of Employment ...... 47 1.3 Physicians’ Place of Employment by Physician Age Group ...... 48 1.4 Other Medical Occupations’ Place of Employment ...... 49 1.5 Distribution of Facility-Shares for Medicare Services ...... 50 1.6 Percent of Total Services and Revenue by Facility-Share ...... 51 1.7 Relative Medicare Reimbursement by Specialty ...... 52 1.8 Distribution of Relative Medicare Reimbursement by Service and Year . . . . . 53 1.9 Physician, Hospital, and Other Ownership of Physician Practices by Specialty . 54 1.10 Relative Medicare Reimbursement and Hospital Ownership by Specialty . . . . 55 1.11 Medicare Patient Share Interactions with Relative Medicare Rates ...... 56 1.12 Specialty-Level Medicare Patient Share Interactions with Relative Medicare Rates 57 1.13 Predicted Hospital Ownership Rates ...... 58 1.14 Predicted Hospital Ownership Rates - Lagged Models ...... 59 1.15 “Pretreatment” Trends in Hospital Ownership ...... 60 1.16 Facility-Share of Outpatient Physician Services Delivered to Medicare Patients: 2012 - 2015 ...... 61 1.17 Proportion of Medicare Providers Who Perform At Least 95% of Outpatient Services in a Hospital: 2012 - 2015 ...... 62
2.1 FCA Referrals, Investigations, and Actions ...... 110 2.2 FCA Settlements and Judgments ...... 111 2.3 State FCA Implementation ...... 112 2.4 State FCA Implementation by Type ...... 113 2.5 State FCA Implementation by Federal Medical Assistance Percentage . . . . . 114 2.6 Medicaid Fraud Control Unit Expenditures ...... 115 2.7 Total Medicaid Spending for All Services ...... 116 2.8 Total Medicaid Spending for Prescription Drugs ...... 117 2.9 Total Number of Medicaid Eligibles ...... 118 2.10 Distribution of Estimated Off-Label Usage Probabilities ...... 119 2.11 Spending and Prescription Volume - Off-Label Prone Drug Classes ...... 120 2.12 Spending and Prescription Volume - Not Off-Label Prone Drug Classes . . . . 121 2.13 Pre-Treatment Prescription Drug Spending ...... 122 2.14 Pre-Treatment Prescription Drug Volume ...... 123 2.15 Dynamic Effects of FCAs - Off-Label Prone Drug Classes ...... 124
vii Chapter 1: Drivers of Physician-Hospital Integration: The Role of Medicare Reimbursement
1.1 Introduction
Physicians are rapidly integrating with hospitals through practice acquisitions and direct employment. This type of vertical integration may lower health care costs by coordinat- ing care between providers or potentially increase prices by limiting competition between physicians and conferring market power to integrated providers. Despite the explosive rate at which physicians are seeking hospital employment and the significant portion of health expenditures devoted to hospital and physician services, there has been little recent work which identifies the factors which are driving physician-hospital integration.1
This paper analyzes the relative importance of factors which drive the integration deci- sions of physicians and hospitals. There is a growing literature which examines the conse- quences of physician-hospital integration on prices and service utilization. While there is no general consensus on the effects of physician-hospital integration, there is increasing evidence that integration decreases competition, increases prices and utilization, and does not im- prove quality of care.2 Few studies focus specifically on consequences in physician markets or analyze which factors drive the integration decision on the extensive margin. Under- standing why physicians are seeking hospital employment through acquisition of physician practices may inform the debate as to how integration may lead to cost and utilization reductions or increases.
I empirically estimate the responsiveness of physicians to Medicare “provider-based” billing policies, which compensate integrated hospital-owned and independent free-standing physician practices differentially. Provider-based billing has long been criticized by gov-
1Spending on physician and hospital services alone accounts for over half of national health expendi- tures or a combined 9.3% of GDP in 2015 (see National Health Expenditure data for 2015 “Hospital” and “Physician and Clinical Services” expenditure categories (CMS, 2016)). 2Gaynor, Mostashari, and Ginsburg (2017) provide a brief discussion of recent consolidation in health care markets, including vertical integration between hospitals and physicians. See Madison (2004), Ciliberto and Dranove (2006), Cuellar and Gertler (2006), Baker, Bundorf, and Kessler (2014), Robinson and Miller (2014), Neprash, Chernew, Hicks, et al. (2015), Capps, Dranove, and Ody (2017), and Koch, Wendling, and Wilson (2017a) which study the effects of physician-hospital integration.
1 ernment agencies and economists as inadvertently encouraging consolidation of physician practices and hospitals, but until recently no study has quantified its effect on organizational structure or examined the impact across multiple physician specialties.3 Recent studies in- dicate that Medicare’s role in driving integration may be heterogeneous across specialties.
Ody and Dranove (2016) find that changes in the relative prices paid for Medicare services provided in integrated versus independent practices account for roughly one-third of the ob- served integration between physicians and hospitals since 2007. On the other hand, Alpert,
Hsi, and Jacobson (2017) find that neither a 2005 payment cut in chemotherapy drugs, nor increased eligibility for the 340B drug program under the Affordable Care Act explained the observed increase in vertical integration of oncology practices from 2003 through 2015.
This paper complements these studies by examining the effect of Medicare reimbursement changes across multiple specialties and allowing responses to differ across physician type.
This paper also documents differential changes in hospital ownership of physician practices across specialties, which has received little attention in previous work.
I link Medicare administrative prices and service utilization and payment data from the Centers for Medicare and Medicaid (CMS) with physician practice characteristics and ownership status from the National Ambulatory Medical Care Survey (NAMCS) for years
2005 through 2015 to empirically estimate the effect of changes in relative reimbursement rates for hospital and physician-owned practices on vertical integration between hospitals and physicians. I utilize a difference-in-differences methodology where annual updates to
Medicare reimbursement rates and differential service utilization across physician specialties result in varying intensity of exposure of each specialty to Medicare’s provider-based billing policy. Identification relies on variation in relative reimbursement rates over time, within a physician specialty, and across specialties since providers in each specialty typically bill for different types of services.
Results indicate that some physician specialties are responsive to increases in relative
Medicare reimbursement rates paid to integrated providers. A 10 percent increase in the relative price paid to integrated providers results in a 1.9 percentage point increase in hos-
3See OIG (1999), OIG (2016), GAO (2015), and Gaynor, Mostashari, and Ginsburg (2017).
2 pital ownership of physician practices in Medical Care specialties (cardiology, dermatology, and neurology). This accounts for about one-third of the observed increase in hospital ownership in these specialties over 2005 through 2015. Allowing for a lag in the response in hospital ownership to changes in relative prices, Medical Care and Surgical Care special- ties (general surgery, orthopedic surgery, ophthalmology, and urology) exhibit an increase in hospital ownership of 1.4 and 2.1 percent in response a 10 percent increase in relative prices. Primary care specialties do not seem responsive to changes in Medicare reimburse- ment. Across all specialties, there is some evidence that physicians who see relatively more
Medicare patients are more responsive to increases in the gap in Medicare reimbursement rates between hospitals and physician offices. Since acquired practices may charge higher rates to Medicare and private insurers without any subsequent changes in where or how health care services are delivered as well as refer patients to the acquiring hospital, there exists little incentive for providers who integrate in response to Medicare’s provider-based billing policy to reduce costs or improve the quality of care. If integration is only a re- sponse to a growing premium in Medicare reimbursement rates for integrated providers, then this may result in an inefficiently high level of vertical integration between physicians and hospitals.
This paper adds to the growing literature addressing vertical integration between hos- pitals and physicians. It is one of only a few recent studies which address the motivation for vertical integration, as opposed to consequences, and shows that, while Medicare pay- ment rules may be an important contributing factor to the recent, rapid growth in the number of hospital-owned practices, it does not seem to explain increased integration for all physician specialties. Specifically, the increase in hospital ownership of Primary Care specialties, including general and family practice, obstetrics and gynecologists, and pedi- atricians, does not appear to be explained by changes in Medicare provider-based billing payment rates. CMS’ recent implementation of “site-neutral” payments which eliminate the provider-based payment differential for some newly acquired physician practices may slow the pace of integration, but we should not expect this to affect all specialties equally.
3 1.2 Background and Literature
1.2.1 Integration in the 1990s
Physician-hospital integration is not a new phenomenon. Growth in the use of managed care organizations (MCOs) and health maintenance organizations (HMOs), which attempt to steer insured patients to preferred lower-cost providers, led to hospital and physician alignment to maintain bargaining leverage in negotiating reimbursement rates with insurers and securing managed care contracts (Burns, Bazzoli, et al., 2000; Cuellar and Gertler,
2006). Partnerships between physicians and hospitals in the 1990s are generally regarded as a failure and most partnerships subsequently dissolved (Burns and Pauly, 2002). Figure
1.1 presents data from the American Hospital Association (AHA) detailing the percent of hospitals engaged in various affiliations with physicians. Each organizational structure represents some level of vertical integration between physicians and hospitals; models such as integrated salary models (ISM) (not shown in the chart) represent the tightest form of vertical integration and include direct hospital employment or ownership of a physician practice. In a physician hospital organization (PHO), the joint entity negotiates payment and contracts, but physicians retain ownership of the group or practice. Management service organizations (MSO) simply provide administrative support to physician practices and may purchase services such as outsourced billing, malpractice insurance, or lease equipment on behalf of the practice. Independent practice associations (IPA) are the weakest form of vertical integration where the organization acts as an intermediary between managed care plans and independent physicians. Physician-hospital affiliations peaked around 1995 and have since declined. Not shown in this chart are integrated salary models or direct employment, which has consistently increased since 1993 and has become the dominant form of physician-hospital vertical integration (Burns, Goldsmith, and Sen, 2013).
1.2.2 Integration Today
There has been a renewed interest in vertical integration between physicians and hos- pitals since the early 2000’s. More recently, the Affordable Care Act established incentives for providers to form Accountable Care Organizations, which bring together providers both
4 horizontally and vertically to coordinate care for chronically ill patients (Burns and Pauly,
2012; Frech et al., 2015). Although many physician-hospital partnerships of the 1990’s dissolved, direct employment and hospital acquisition of physician practices has increased rapidly.4
Figure 1.2 shows the place of employment for physicians and surgeons (henceforth re- ferred to as “physicians”) from 2005 through 2015, as defined by industry and occupation codes in the American Community Survey (ACS).5 Trends in Figure 1.2 are consistent with other reported estimates and show a consistent decrease in physician office employment and simultaneous increase in hospital employment of physicians. From 2005 to 2015, the percent of physicians working in a physician office declined from just under 50 percent to about 40 percent (a relative decrease of 19.4 percent). Hospital employment increased by a relative
20.8 percent, surpassing physician office employment in 2009.
Figure 1.3 further breaks down physician employment by age group. If recent medi- cal school graduates today value a better balance of work and leisure and choose to forgo higher incomes for less administrative responsibility by seeking hospital employment, then the trend of increasing hospital employment may be driven by the changing preferences of younger physicians relative to older physicians (Kocher and Sahni, 2011). Figure 1.3 shows that younger physicians, age 25 to 44, are much more likely to work in a hos- pital setting across all years, but the trend of increasing hospital employment does not seem to have differentially affected this group, relative to older physicians. There is also no evidence that older physicians, over 65, are increasingly looking to sell their practice
4Kocher and Sahni (2011) present data from the Medical Group Management Association showing physi- cian ownership of practices decreasing from about 70% of practices in 2002 to less than 50% in 2008. Hospital ownership increased from just over 20% to over 50% over the same time period. Kane and Emmons (2013) find that, according to the American Medical Association’s 2012 Physician Practice Benchmark Survey, 60% of physicians work in practices wholly owned by physicians, compared to 23% in practices at least partly owned by a hospital, and an 5.6% that are directly employed by hospitals. Subsequent AMA surveys show hospital ownership increasing to 25.6 and 25.4 percent and direct hospital employment to 7.2 and 7.4 percent in 2014 and 2016 respectively (Kane, 2015; Kane, 2017). Using SK&A data Richards, Nikpay, and Graves, 2016 find that the independent physician practice share declined from 73 to 60 percent from 2009 to 2015, and the share of practices integrated with a health system increased from 7 to 25 percent over the same period. 5The shift from physician office to hospital place-of-employment also represents a shift from ownership to employee status. Physicians who report working in physician offices are predominantly self-employed, with about 54 percent reporting self employment versus 5 percent of those working in a hospital.
5 or seek hospital employment. Across all age groups, physicians are increasingly likely to seek hospital employment.6 Although this is only suggestive evidence, it does not appear that changing physician preferences or demographics are driving the overall trend towards physician-hospital integration.
If the trend of increasing hospital employment is present in other medical occupations besides physicians and surgeons, then we may suspect that more general industry trends are pushing all medical professionals into hospitals. Hospital market concentration has increased gradually over the last decade, and hospitals may seek more employees to differ- entiate themselves from competitors, gain market power, or take advantage of economies of scale and coordinate care from numerous providers and settings (Gaynor and Town,
2011; Cutler and Scott Morton, 2013). As a comparison of physician employment to other medical occupations, Figure 1.4 shows the proportion of nurses (RN’s and LPN’s) and physician assistants employed by hospitals and physician offices from 2005 to 2015. Over- all, place-of-employment trends for other medical occupations show little change over this time period. Therefore the trend of increasing hospital employment and decreasing physi- cian office employment appears unique to physicians and provides evidence that there is some fundamental difference in physician markets versus other health care providers. It is important to understand what factors are rapidly pushing physicians from self-employment in physician practices to employment in hospital settings.
1.2.3 Consequences and Causes of Physician-Hospital Integration
Consequences of Vertical Integration
Empirical literature on the consequences of vertical integration between physicians and hospitals has grown tremendously since 2006. Studies typically examine the impact of in- tegration on prices and utilization, with less attention given to heath care quality. Until recently, most studies focused on hospital, rather than physician markets. Earlier stud- ies were inconclusive on the net effects of integration. Using similar categorizations of
6Grouping physicians into five 10-year age groups yields similar trends as presented in Figure 1.3.
6 physician-hospital integration, Cuellar and Gertler (2006) found that integration led to higher hospital prices with no reduction in cost or quality improvements, but Ciliberto and
Dranove (2006) found no evidence of hospital price increases over a similar time period.7
More recently, Baker, Bundorf, and Kessler (2014) associate “tight” vertical integration of
physicians and hospitals with higher prices and total spending for Medicare beneficiaries
at the county level.8 Robinson and Miller (2014) find that local and multi-hospital-owned physician organizations in California incur 10 to 20 percent higher expenditures per patient, respectively, compared to physician-owned organizations.
There has been almost no discussion of whether not-for-profit and for-profit hospitals differ in either the incidence or outcomes of vertical integration with physicians. As Duggan
(2000) points out, for-profit and not-for-profit hospitals do not necessarily differ in their response to financial incentives. Cuellar and Gertler (2006) briefly discuss for versus not-for- profit status and integration with physicians and find that integration for both types results in higher prices for managed care patients, yet for-profit hospitals have slightly higher prices for indemnity patients. Integration is common for large, not-for-profit teaching hospitals, and integration for these hospitals does not seem to result in higher prices or lower costs, but does improve quality outcomes for patients.9
To date, studies examining potential quality improvements in health care services due to integration between hospitals and physicians have not found strong evidence of gains in quality. Madison, 2004 examines the effect of physician-hospital integration on Medi- care beneficiaries admitted to a hospital with acute myocardial infarction (AMI). Although integrated organizations do result in a higher intensity of treatment, there is no measur- able reduction in mortality or readmissions. Cuellar and Gertler (2006) examine inpatient mortality and measures of overuse and patient safety and find that only some forms of integration lead to quality improvements. Tightly integrated, “Fully Integrated Organiza-
7Cuellar and Gertler (2006) analyze hospital data from Arizona, Florida, and Wisconsin whereas Cilib- erto and Dranove (2006) analyze California data. 8Baker, Bundorf, and Kessler (2014) use categorizations of physician-hospital integration similar to AHA categories. Ciliberto and Dranove (2006) and Cuellar and Gertler (2006) use these measures as well. 9There are no for-profit “Fully Integrated Organizations" in the data; it is difficult to make any distinction between this type of integration and for versus not-for-profit status.
7 tions” are the only organizational form associated with mortality reductions, however, this form of organization is also comprised mostly of large, nonprofit academic and teaching hospitals. Most integrated organizational forms had no statistically significant relationship with any of the quality measures examined. More recently, Bishop et al. (2016) find that physician practices that were acquired by a hospital increased their use of “care manage- ment processes," including disease registries, nurse care managers, reporting quality data to physicians, patient reminders, and patient education. Although care processes may be changed as a result of integration, there has been little evidence showing that any effect carries through to patient outcomes. Scott et al. (2017) examine hospitals who switch from physicians with privileges to direct employment and find no effects on quality as measured by hospital mortality rates, 30-day readmission, length of stay, and patient satisfaction scores. In a working paper, Koch, Wendling, and Wilson (2017b) find no consistent or siz- able quality improvements subsequent to hospital acquisition across a range of diabetes and hypertension health outcomes and indicators. Combined, the few studies which examine quality and outcomes indicate that any observed increase in price is likely not caused by changes in the underlying quality of care. As a side note, however, all studies which have examined quality outcomes of physician-hospital integration have measured integration as reported at the hospital level with AHA data. It could be that changes in integration status of the hospital only affect a small proportion of total physician employees and that the or- ganizational change is not large enough to affect quality outcomes measured at the hospital level. No study has quantified the effect of hospital ownership or employment on quality outcomes using data on individual physician-level ownership status.10
More closely related to this paper are empirical studies which have examined the conse- quences of physician-hospital integration in physician markets. Neprash, Chernew, Hicks, et al. (2015) find that prices paid by commercially insured patients for outpatient services increased by 3.1 percent as vertical integration between hospitals and physicians increased,
10Baker, Bundorf, Devlin, et al. (2016) compare American Hospital Association data and SK&A market- ing data, both of which measure hospital ownership and employment of physicians, but each using different methods. AHA data measures ownership from the hospital’s perspective, and cannot be used to identify individual, hospital-owned physicians. SK&A data is based on a sample of office-based physicians and measures hospital-ownership from the physician’s perspective.
8 with no significant effect on service utilization. Capps, Dranove, and Ody (2017) find that vertical integration of physicians leads to about a 14 percent increase in prices, and almost half of the estimated price increase is attributable to Medicare’s provider-based billing pol- icy which allows integrated physician offices to bill Medicare at higher rates and charge additional “facility fees.” Estimated price increases vary across specialties, with prices for services provided by primary care physicians increasing by 12 percent and cardiologist ser- vices increasing by over 34 percent post integration. Anesthesia and diagnostic radiology specialties exhibit only a small and insignificant increase in prices after vertical integration with a hospital. This paper is the only study which highlights the heterogeneous effects of vertical integration across physician specialties. If price increases due to vertical inte- gration vary by physician specialty, then perhaps the motivation for integrating also varies by specialty. The authors do point out that they see no obvious link between specialties with higher expected price increases due to vertical integration and the share of spending by vertically integrated physicians, except for cardiology.11
Koch, Wendling, and Wilson (2017a) perform a post-merger analysis of acquired physi- cian practices using Medicare claims data and industry reports on acquisitions. Three facets of physician behavior are measured: physician billing in the acquired office, physician billing in the acquiring hospital, and aggregate acquiring hospital system billing. They find that acquired physicians bill roughly 70 percent less in terms of claims and total spending in an office setting. Those same physicians bill for relatively more work in the acquiring hospital, with total claims increasing over 50 percent on average with a similar magnitude effect on spending. This could represent a true change in the location where the services are performed, pre- versus post-merger, or this could represent a change in how the services are billed, with acquired practices billing as provider-based departments. There is some evidence that acquired physicians bill less at rival, non-acquiring hospitals after a merger.
Overall, acquiring hospitals see a relatively smaller change post-merger, with total claims increasing by less than 10 percent and an insignificant 2 to 3 percent increase in total spend- ing. This could be because of the small fraction of total doctors at the acquiring hospital
11See Capps, Dranove, and Ody (2017) Figure 3, pg. 35.
9 made up by acquired doctors, variation in the number of affected doctors across mergers, and variation in individual hospital use among acquired physicians in multi-hospital sys- tems. It is estimated that total spending increases by 18 percent on net. These results are consistent with acquiring physicians and hospitals responding to Medicare’s provider-based billing policies. Acquired physicians shift billing to hospital outpatient departments, espe- cially for evaluation and management “clinical” services where Medicare payment policy most likely overcompensates hospitals relative to independent physician offices.
Causes of Vertical Integration
Less attention has been devoted to identifying the motivating factors which have con- tributed to the recent increase in integration between physicians and hospitals. Brunt and
Bowblis (2014) find that increases in insurance market concentration result in increased hospital ownership of primary care practices. McCarthy and Huang (2016) also find that physicians align with hospitals in response to increased insurance market concentration and that hospitals seek vertical integration with physicians in more competitive hospital mar- kets. Although there is potential for hospitals and physicians to integrate in response to
ACO’s formed by the Accountable Care Act, Neprash, Chernew, and McWilliams (2017)
find no evidence that physician-hospital integration increased in markets with higher ACO penetration. Instead, consolidation was already increasing prior to the passage of the ACA and establishment of the ACO program.
Only recently has any study examined Medicare provider-based reimbursement as a potential factor driving integration on the extensive margin. Song et al. (2015) analyze
Medicare and private insurance claims data and focus on a select group of cardiology services where Medicare reimbursement rates were cut for independent physician offices relative to hospital outpatient rates.12 They find that after the reimbursement cuts, billing volume shifts to hospital outpatient settings at an increasing rate for both Medicare and privately insured patients. They interpret this shift in volume as evidence of both increasing vertical
12Song et al. (2015) analyze the share of HOPD billing volume of myocardial perfusion imaging (MPI), echocardiograms, and electrocardiograms.
10 integration of cardiologists and a change in the setting where services are provided. This study cannot rule out other factors that may be shifting health care services to outpatient settings, such as the general decline in demand for inpatient services, and results may not be causal or generalizable to other specialties. Alpert, Hsi, and Jacobson (2017) examine the impact of both a 2005 Medicare payment reform, which reduced reimbursement for chemotherapy drugs, and the 2010 ACA eligibility rule changes for the 340B Drug Discount
Program on vertical integration of oncologists.13 The authors find that neither of these policy changes explain the rapid increase in physician-hospital integration in oncology, which doubled from 30 to 60 percent over the 2003 through 2015 period. The authors conclude that the rapid growth in vertical integration in oncology markets, largely accelerating since
2010, may be due to other post-ACA factors such as bundled payments, incentive to form
ACOs, increasingly complex payment design, and/or increased administrative burden such as electronic health record requirements.
The only study to examine the effect of Medicare payment policy on vertical integra- tion across multiple physician specialties is Ody and Dranove (2016). The authors use a
2010 change in how Medicare reimbursement rates are calculated to identify a causal rela- tionship between a measure of the relative price of site-specific reimbursement, physician employment, and increasing utilization of hospital outpatient departments for services that could be delivered in a physician office setting. Overall, Ody and Dranove (2016) find that the 2010 update to Medicare payment rules explains 20 percent of the observed increase in physician hospital integration from 2010 to 2013. Although, in their companion paper,
Capps, Dranove, and Ody (2017) show that changes in price due to vertical integration and the ability to charge additional “facility fees” varies substantially by physician specialty,
Ody and Dranove (2016) do not explore variation in response to Medicare payment rates across specialties.
13As Alpert, Hsi, and Jacobson (2017) explain, both of these policies should potentially result in increased vertical integration between oncologists and hospitals. Payment cuts affecting physician practices could be mitigated by integrating with a hospital and leveraging a subsequent increase in market power to negotiate higher discounts from pharmaceutical companies. The 340B Drug Discount program provides discounts to certain hospitals when purchasing drugs. Newly eligible hospitals may wish take advantage of lower drug costs (and increased profit margins) by expanding their patient base through integration with oncology practices.
11 1.3 Medicare Reimbursement
Medicare reimbursement policies have been criticized for inadvertently encouraging hos- pital acquisition of physician practices and physician employment by paying higher “facility fees” to hospital outpatient departments who provide identical services as free-standing physician offices (Gaynor, Mostashari, and Ginsburg, 2017).14 Government advisory agen- cies have also expressed concern over Medicare payment policies’ adverse effects including health care provider consolidation, increased service utilization, and increased total spend- ing (MedPAC, 2012; MedPAC, 2013; GAO, 2015). A 2012 Medicare Payment Advisory
Commission (MedPAC) report estimates that 20 percent of the growth in outpatient ser- vice volume, which increased by 28 percent from 2004 through 2010, is due to an increase in outpatient department evaluation and management “office” visits. The report notes that this increase could be driven by hospital acquisition of physician practices. The commis- sion recommended that payment rates for evaluation and management visits be equalized across payment settings to remove the financial incentive for hospitals to purchase physician practices and shift service billing to the more costly outpatient setting. As an example, the report shows that in 2011 a midlevel office visit was over 80 percent more expensive in an outpatient facility versus a free-standing physician practice. (MedPAC, 2012).
Medicare rates are relevant not only to providers who see Medicare patients, but the privately insured as well. Clemens and Gottlieb (2017) show that private insurance reim- bursement rates follow Medicare prices, on average about 16 percent higher, and that the extent of price following is related to physician market competition relative to insurers.
Price following behavior may be due to mechanical benchmarking of private insurance rates
14Facilities that are determined to have “provider-based” status, as defined by 42 CFR §413.65, are reimbursed under the hospital Outpatient Prospective Payment System. Requirements for provider-based status include the facility operating under the same license as the main provider, professional staff of the facility have clinical privileges at the main provider facility, medical records for patients treated at the facility are integrated in a unified system of the main provider, patients who require further care have full access to all services of the main provider and are referred when appropriate. “Off-campus” facilities are additionally required to be 100 percent owned by the main provider, share a governing body, and be within a 35-mile radius of the main provider’s facility. See https://www.law.cornell.edu/cfr/text/42/413. 65. Beginning January 1, 2017, off-campus provider-based departments which provide outpatient services are no longer reimbursed under the OPPS, unless they were established prior to November 2, 2015 and have continuously billed for covered outpatient services from the same location. See https://www.cms.gov/ Newsroom/MediaReleaseDatabase/Fact-sheets/2016-Fact-sheets-items/2016-11-01-3.html.
12 to Medicare rates (i.e. a rate of 120% of Medicare) to reduce negotiation costs, or it may reflect providers’ outside option of serving Medicare patients. Consistent with a bargaining power framework, price following is less intense in markets where provider concentration is high relative to insurance market concentration. If vertical integration increases provider bargaining power, then payment differentials that overcompensate integrated practices could be exacerbated as providers demand higher reimbursement rates.
Empirical work analyzing the consequences of hospital acquisition of physician practices supports the relationship between Medicare’s administrative prices and private insurance reimbursement rates. Ody and Dranove (2016) find evidence that a 2010 update to prices in Medicare’s physician fee schedule impacts private insurance rates for non-facility services
(with a coefficient of about .5, indicating that private rates are roughly twice as high as
Medicare rates). Private insurers, therefore, do set prices that reflect Medicare’s provider- based billing policy.15
Medicare’s provider-based billing policies combined with differences in how reimburse- ment is determined across payment systems and federal regulation including the anti- kickback statute and the Stark Act’s anti-self-referral laws work together to create a wedge in the prices paid for services furnished in freestanding independent physician offices ver- sus hospital-owned offices, which may bill as a hospital outpatient care setting. Higher
Medicare reimbursement paid for outpatient services is meant to compensate a hospital for the additional overhead costs associated with providing care, but these payment poli- cies could overcompensate hospital-acquired physician practices, which may treat the same patients, provide the same services, and remain in the same location as they did as an independent practice, yet which may bill higher rates as an outpatient clinic. Integration between hospitals and physicians in response to payment rate distortions seems contrary to efficiency-improving motivations for integration. To the extent that the differential in
15Ody and Dranove (2016) also find that utilization of services in a facility setting for Medicare and privately insured patients are extremely similar, with the exception of evaluation and management visits. They verify with their claims data provider that E&M visits are not reimbursed in a facility setting, reflecting a rejection of Medicare’s provider-based billing policy for this specific set of services. It is unclear whether this is unique to this data provider, or if private insurers are generally able to choose which services are reimbursed at higher facility-setting rates and which they are unwilling to do so.
13 Medicare reimbursement drives physician-hospital integration, we should not be optimistic that integration will lead to cost reductions, decreases in utilization, or any improvements in quality of care.
1.3.1 Medicare Payment Systems
The two relevant Medicare payment systems for Medicare Part B services are the Medi- care Physician Fee Schedule (MPFS) and the Hospital Outpatient Prospective Payment
System (HOPPS). The MPFS is used to compensate physicians, and other Part B providers, who provide a service for a Medicare patient in any care setting.16 In 2014, payments under the MPFS made up 16% ($69 billion) of all Medicare fee-for-service expenditure (MedPAC,
2016). The MPFS lists payment rates for each unique service for which a provider may bill
Medicare, with approximately 7,400 distinct services.17 Payment rules are updated quar- terly by CMS with the most significant changes occurring in the beginning of the year.18
Further, The MPFS compensates “facility” and “non-facility” providers differently. Non-
facility rates apply to providers in free-standing physician office settings and are typically
higher than facility rates, which apply to services furnished within a hospital. Non-facility
rates are higher than facility rates since they are meant to cover overhead costs of main-
taining an office. Differences in these rates, however, are typically small, and in many cases
the facility and non-facility rates are equal.
The HOPPS payment system compensates hospital outpatient departments for each
service provided, defined by a HCPCS code, with services grouped into Ambulatory Pay-
ment Classifications (APC’s) where reimbursement for each service within an APC is the
16Settings include a physician office, hospital, ambulatory surgical center, skilled nursing facility, other post-acute care facilities, hospices, outpatient dialysis facilities, clinical laboratories, and in-home care. 17Unique services are defined as 5-digit Healthcare Common Procedure Coding System (HCPCS) codes, which are comparable to the American Medical Association’s Current Procedural Terminology (CPT) codes. 18MPFS payments are based upon “relative value units” (RVUs) and include three categories of physician practice costs, (1) work RVUs, (2) practice expense RVUs, and (3) malpractice RVUs. Work RVUs measure the time and intensity required of a physician to provide a service, practice RVUs cover the cost of maintaining an office such as office rent, supplies, staff, and other overhead costs. Malpractice RVUs measure the cost of malpractice insurance. Each RVU category is further adjusted with a geographic practice cost index (GPCI) adjustment factor (CMS, 2016 “Medicare Physician Fee Schedule - Payment System Fact Sheet Series”).
14 same flat rate.19 This payment, sometimes referred to as a “facility-fee,” is separate from reimbursement paid to a physician if he or she provides services in an outpatient setting.20
The HOPPS payment is usually much larger than the difference in facility and non-facility rates paid to physicians and results in total payments to hospital outpatient departments that are much higher than payments made to free-standing physicians offices for the same service (MedPAC, 2011).
1.3.2 Federal Anti-fraud and Abuse Regulations
Federal anti-kickback laws and the Stark anti-self-referral law prevent hospitals and physicians from engaging in financial relationships based on referrals or service utilization.
Without anti-kickback laws in place, an independent physician could refer patients to a hospital outpatient department and receive a cut of the higher payment rate as a bonus for his or her referrals. Stark and anti-kickback laws prevent independent physician prac- tices from sharing in higher payment rates to hospitals generated by provider-based billing.
Notably, however, both Stark laws and the anti-kickback statute provide an exception for hospitals which directly employ physicians.21 While it is illegal for a hospital employer to pay outright for referrals from an acquired physician, it may be able to offer a higher salary since many services are valued more highly by Medicare when billed in a hospital outpatient department. A 2010 Medical Group Management Association survey shows that although physicians in non-hospital-owned practices have a higher median income, the median in- come per work relative value unit (RVU) is about 7 percent higher for specialists, indicating that hospital owned physicians are paid more conditional on their productivity.22 Physician practices that are hospital owned may be pushed to self-refer to the owning hospital and
19HOPPS payment rates are adjusted by a geographic cost index, similar to MPFS adjustments. The adjustment factor is based upon the CMS’s hospital wage index and applies to 60% of the payment rate. Hospitals may also receive additional payments for new technologies, costly outlier services, and specific payments to cancer centers and children’s hospitals. 20I refrain from using the term “facility-fee,” although this is the terminology used by Capps, Dranove, and Ody (2017) and Ody and Dranove (2016), to avoid confusion between the HOPPS payment and facility versus non-facility rates in the MPFS. 21See Code of Federal Regulations Title 42 Ch. 4 §411.357. Employers must pay physicians for identifiable services at a fair market rate and may not take into account volume or value of any referrals. 22see https://www.mgma.com/blog/highlights-of-mgma-s-2010-physician-compensation-survey
15 increase the volume of services for which higher reimbursement applies. Vertical integra- tion between physicians and hospitals is one way which providers can take advantage of differential Medicare payments.
1.4 Data
I combine data on Medicare administrative prices from annual updates of the MPFS relative value files and HOPPS payment rates, utilization volume and payment data from
CMS Medicare Utilization Data used in updating the Medicare Physician Fee Schedule, and physician practice characteristics from the National Ambulatory Medical Care Survey
(NAMCS) for years 2005 through 2015 to measure the impact of Medicare payment rate changes on physician-hospital integration.
1.4.1 Medicare Physician Fee Schedule and Hospital Outpatient Prospective
Payment System
I use MPFS relative value files from 2005 through 2015 to construct procedure-level administrative payment rates for services provided by physicians in offices and outpatient hospital-based departments.23 Physician office rates are constructed using the non-facility payment rate, and hospital outpatient-department rates are constructed using the facility rate. Services in the MPFS are linked to the Hospital Outpatient Prospective Payment
System by HCPCS code in each year.
Equations 1.1 and 1.2 below summarize how the reimbursement rate for service j in year t is calculated for both an independent, physician-office-based provider (“PO”) and an integrated physician office billing as a hospital outpatient department (“HOPD”):
PO N Payjt = MPFSjt (1.1)
23See https://www.cms.gov/Medicare/Medicare-Fee-for-Service-Payment/PhysicianFeeSched/ PFS-Relative-Value-Files.html. I use annual January updates of the MPFS RVU files and HOPPS payment rates to construct the relative reimbursement rates. If rates for January are revised, I use the latest revision.
16 HOPD F Payjt = MPFSjt + OPPSjt (1.2)
N represents the non-facility MPFS rate and F represents the facility rate. The OPPS payment rate is determined at the service-level and is paid only to providers billing in a hospital outpatient setting.24,25 Since payment rates are defined at the HCPCS code level, and codes are sometimes added, modified, or deleted with each annual update, CMS HCPCS code crosswalks are used to ensure that comparisons of services across years are valid.
1.4.2 Medicare Utilization and Payment Data
Medicare payment rates are calculated for each service in the MPFS. Physicians are differentially exposed to disparities in HOPD and physician office payment rates depending upon the types of services for which that physician bills. Since I do not have claims-level data and the NAMCS data does not contain information on HCPCS/CPT codes billed for each physician-patient encounter, I construct an aggregate measure of service utilization and spending which reflects the various types of services most commonly billed by each physician specialty in the NAMCS data. CMS provides utilization “crosswalk” data for select years included with quarterly updates to the Medicare Physician Fee Schedule. Aggregate service volume and charges are reported by specialty, year, and HCPCS code.26
I aggregate payment and utilization by service and specialty to construct an average
“basket” of services that each specialty typically bills Medicare. Since each specialty varies in the types of services provided and provider-based billing differences in HOPD and physi- cian office payment rates vary at the service level, I am able to calculate a weighted average
HOPD-to-PO ratio of reimbursement rates for each specialty. The weighted average HOPD- to-PO ratio, which I will refer to as a “relative price” reflects the gap in Medicare payment
24National-level payment rates which do not include geographic adjustments are used in calculations. 25The MPFS includes separate professional and technical components for many services, denoted with indicators “26” and “TC”. These two components sum to equal the total reimbursement amount for a given service. For reimbursement paid for services provided in a facility setting, the physician only receives the professional fee component. For services with separate professional and technical components, I include only the professional component for calculated HOPD reimbursement rates. 26I use CMS “crosswalk” utilization data for years 2003, and 2007 through 2015. 2003 data are used to estimate weights for year 2005, and 2007 utilization data are used in 2006.
17 rates for hospital-owned physician practices billing as outpatient departments versus inde- pendent physician-owned practices for a group of services commonly billed by physicians in a particular specialty. The relative price measures the multiple of revenue on Medicare services that would be generated if an independent physician practicing in an office were to shift all of his or her billing to a hospital outpatient department setting, but maintain an identical service mix.
Service utilization weights, which are used to calculate the weighted-average relative price, are calculated separately for each HCPCS code, specialty, and year. Specialties in- clude cardiology, dermatology, family practice, general practice, general surgery, internal medicine, neurology, obstetrics/gynecology, opthalmology, orthopedic surgery, otolaryngol- ogy, pediatrics, psychiatry, and urology. Each weight is calculated as proportion of allowed charges for each service provided in an office setting.27 I attempt and limit the sample of
services to those that may be performed both in a physician office and hospital setting. To
do this, I calculate a facility-share for each service which represents the amount of revenue
generated for that service in a hospital (facility) setting. I exclude the top and bottom
1 percent (services performed in a facility or hospital setting 99 percent of the time or
greater or less than 1 percent of the time). Figure 1.5 shows the distribution of service-level
facility-shares. There are two clear bunching points at zero and one where some services
are almost never performed across facility and non-facility settings. Limiting the sample
to HCPCS codes with a facility share between 1 and 99 percent leaves about 54 percent of
services in the sample, and 71 percent of charges (see Figure 1.6). Limiting the sample to
services with a facility-share between 5 and 95 percent leaves 33 percent of services which
comprise 29 percent of total charges. In calculating the relative prices and utilization of
Medicare services, I limit the sample to services with a facility-share between 1 and 99
percent. Omitting these services does not greatly affect the final weighted-average relative
price.
The CMS utilization data have a couple of limitations: First, utilization and payments
27The average of the Medicare allowed amount for any given service is the sum of the amount Medicare pays, the deductible and coinsurance amounts that the beneficiary is responsible for paying, and any amount that a third party is responsible for paying.
18 are reported only for Medicare beneficiaries and may not be generalizable to all patients served by a particular practice or specialty. If physicians bill for a substantially different mix of services for non-Medicare patients and these patients make up proportionally more of the physician’s practice, then the calculated weights for each service may not represent the true mix of services performed by the physician. This would induce measurement error in the weighted average relative price. In their analysis of a 2010 cost survey update and MPFS pricing, Ody and Dranove (2016) compare Medicare services performed in both facility and office settings to services performed for privately insured and find that the setting of care for all major procedures is highly correlated across both patient types.
Second, CMS utilization data are not available for all years in the NAMCS sample window, and in years where the data are available the HCPCS codes must be crosswalked
“backwards” so that they are consistent with code values listed in the corresponding year of the MPFS and HOPPS. CMS provides HCPCS crosswalks which link codes through updates over time, however, these files are only available from the 2009-2010 HCPCS code update and forward, and they are sometimes incomplete or contain errors. Due to these limitations, it is not always possible to match a HCPCS code and its estimated utilization weight to a corresponding service in the MPFS and HOPPS. In the case where HCPCS codes are not matched, estimated utilization weights are constrained to sum to one within each specialty and year. This implicitly assumes that the weighted-average relative price for unmatched services is similar to that of matched services.
Table 1.1 summarizes changes in weighted-average reimbursement rates for a freestand- ing physician office (PO), an integrated hospital-owned physician practice (HOPD), and relative price between 2005 and 2015.28 Cardiology, urology, and orthopedic surgeons wit- ness the largest increase in the gap between HOPD and PO reimbursement, with the relative price increasing by 47 percent, 34 percent, and 30 percent respectively. The gap between
HOPD and PO reimbursement increases only slightly for dermatology, general and family practice, and internal medicine, and decreases by 11 percent for psychiatry. Figure 1.7
28The basket of procedures included are based upon 2003 and 2007 through 2015 utilization aggregates from CMS. Prices reflect annual updates to these procedures in the MPFS and HOPPS payment systems.
19 shows graphically for each specialty how the relative price paid to integrated providers changes over time. The light-gray lines track changes in the relative price for all specialties in the NAMCS data, and the specialty of interest is bold and in blue. Figure 1.7 makes it clear that some specialties have been more greatly exposed to price differentials resulting from Medicare’s provider-based billing rules. I utilize variation in this “treatment” within specialty and across time to identify the impact of Medicare payment rules on integration.
Increases in the weighted-average relative price could be caused by changes either in reimbursement rates as CMS updates payments in the MPFS and HOPPS systems, or it could be due to a shift in utilization (in the office setting) towards services where the relative reimbursement rate is higher for providers in the HOPD setting. To see which factor is driving changes in the aggregate HOPD-to-PO measure, Figure 1.8 shows the distribution of relative reimbursement rates by service for 2005, 2008, 2011, and 2014. This figure shows that the gap in reimbursement between HOPD and PO settings is increasing over time.
There is a bunching of services where the relative reimbursement is between one and two, but over time the distribution shifts left, indicating an increase in HOPD reimbursement relative to PO reimbursement.29
1.4.3 National Ambulatory Medical Care Survey
NAMCS data from 2005 through 2015 provides information on physician practice char- acteristics including ownership status, physician specialty, solo versus group ownership, and patient visit characteristics such as patient age, sex, and race, diagnoses, and some types of services provided. Physicians included in the survey are limited to non-federally employed office-based physicians primarily engaged in patient care in freestanding outpatient clinics and physician offices. Excluded from the survey are physicians in the specialties of anes- thesiology, pathology, or radiology, and physicians directly employed in hospital outpatient
29This could reflect true reimbursement differentials or changes in how services are bundled in the HOPPS system. For some services that are frequently performed together, payment for both will be bundled into the HOPPS payment rate for only one of the services. Bundling is less common in the MPFS, and so if a service is bundled in the HOPPS the ratio of HOPD-to-PO reimbursement will overstate the payment difference for that service, but understate it for the other bundled services where there is no separate HOPPS payment rate.
20 departments, emergency departments, and ambulatory surgery centers (which are included in the NHAMCS survey). Therefore, the NAMCS data may not capture the entire shift of physicians from independent practice to hospital employment and will likely underestimate hospital ownership rates. The NAMCS also does not track individual physicians over time, making it impossible to follow a particular physician before and after acquisition; I am only able to calculate the average probability of hospital ownership for an annual sample of physicians and compare changes in the probability of ownership over time and specialties.
Analysis of ACS data (Figures 1.2 through 1.4) reveals a striking trend of increasing hos- pital employment and decreasing physician-office employment among physicians. The shift away from independent practice surely varies across physician specialties who face different input costs, differentially rely on hospital resources and affiliations, and who provide dif- ferent medical services and treat potentially very different patient populations. Data from the NAMCS allows us to analyze trends in the integration status of physician practices by physician specialty.
Figure 1.9 presents the composition of physician practice ownership by specialty.30 The majority of practices across all specialties are owned by a physician or physician group, yet ownership composition varies across specialties. Figure 1.9 ranks specialties by share of non- physician ownership. General/family practices, psychiatry, and pediatrics have the highest rates of hospital/other ownership with 25.0 percent, 24.1, and 20.7 percent of practices owned by a non-physician entity, respectively. Ophthalmologists, urologists, and dermatol- ogists have the lowest shares of non-physician ownership (7.5 percent, 8.7 percent, and 9.6 percent respectively).
Differences in the overall organizational ownership structure across specialties may re-
flect differences in costs of practice ownership incurred by administrative or non-physician labor costs; medical malpractice liability; costs of technology such as imaging devices or
30Statistics calculated with NAMCS data measure the proportion of physician practices organized under each respective ownership type, as opposed to physician employment statistics in the ACS. The percentage of physicians employed by hospitals is not directly comparable to the percentage of physician practices owned by hospitals. Employment statistics may show a larger proportion of hospital ownership/employment since hospitals tend to acquire larger physician practices (Burns, Goldsmith, and Sen, 2013) and physicians such as hospitalists and laborists are included in the measure.
21 adopting electronic health record-keeping; economies of scale that may potentially be re- alized by large physician owned group practices; differences in types of services provided; and hospital incentives which encourage the acquisition of certain specialties over others.
Primary care specialties, such as general practice, family practice, internal medicine, and pediatrics, offer office-based “cognitive” evaluation and management services which are not capital-intensive, and typically do not offer diagnostic and treatment services within their own facility. Medical specialties and surgical specialties offer more intensive procedures which utilize practice-owned diagnostic equipment, and facilities (Hough, Liu, and Gans,
2010).
Fundamental differences in the production of health services by various physician spe- cialties may explain the “baseline” level of physician practice ownership within a specialty, however, they likely do not explain within-specialty changes in ownership structure. Figure
1.10 depicts changes in hospital ownership by physician specialty for years 2005 through
2015, in addition to the relative price measure. Cardiology, general surgery, neurology, and urology exhibit the largest increase in hospital ownership over the sample period. Many of these specialties also have experienced a large increase in the relative price, indicating an increase in the reimbursement premium associated with integrating. Cardiology practices shift away from physician ownership at a striking rate beginning in 2009-2010, increasing from 5.5 percent in 2005-2006 to 37.7 percent in 2012-2014 (an increase of 586%). This shift in ownership coincides with 2010 Medicare fee-cuts for cardiologists practicing in an office setting. Other specialties have not exhibited much change in ownership, and the relative price has remained flat (see dermatology, general and family practice, and pediatrics).
Table 1.2 presents summary statistics by ownership type for physician practices in the
NAMCS data. Physician owned practices are more likely to be solo practices, locate in an
MSA, and see more privately insured patients than hospital owned practices. Physicians in hospital owned practices refer more patients out to other physicians. Table 1.3 shows summary statistics by physician specialty. General and family practice specialties are less represented in physician owned offices than other ownership types. Most specialties are comprised of about one-third solo practices, except for dermatology and psychiatry which have relatively more solo practices. Cardiologists, urologists, and ophthalmologists are much
22 more likely to see older and more Medicare patients.
1.5 Model and Methodology
I construct service-level reimbursement rates for a free-standing independent physician office versus an integrated provider billing as an outpatient clinic. These rates vary over time as CMS updates payment rules annually and as relative weights of each service change over time. I estimate the average weight for each service using physician’s total Medicare revenue generated in an office setting. Weights vary at the specialty-year level and are estimated using the CMS crosswalk utilization data. The weighted average of the relative differential in reimbursement paid to integrated versus independent providers represents the average potential gain in Medicare reimbursement that would be realized if an independent physician integrated with a hospital and shifted all of her billing to an outpatient department setting. The relative exposure to Medicare reimbursement differentials varies by specialty due to the difference in services typically provided.
The ratio of the two reimbursement rates for each service is constructed as:
F HOPD HOPDjt MPFSjt + OPPSjt Payjt = N = PO (1.3) POjt MPFSjt Payjt and the weighted average relative price is then:
J X HOPD R = ω jt (1.4) st jst PO j=1 jt for service j and year t, where ωjst varies by service, year, and physician specialty, and sums to one in each year for each specialty. Some procedures are not associated with a price in the
MPFS for various reasons, and so these services do not contribute to the weighted average reimbursement measure.31 Services where a ratio of HOPD to PO reimbursement can be calculated typically make up 70 to 90 percent of revenue in the CMS Medicare utilization data.
31Some prices are not included in the MPFS because they are determined by the carrier, are not reim- bursed under the MPFS, or the code is otherwise excluded or deleted.
23 I estimate the effect of Medicare reimbursement policy on hospital ownership of physician practices using a model of the following form:
Yist = βRst + γXit + αs + δt + εist (1.5)
where Yist is a binary indicator for hospital ownership of physician practice i, of specialty s, in year t, Rst is the weighted average relative Medicare reimbursement rate paid to provider-based hospital outpatient departments, Xit includes physician practice and pa- tient characteristics such as MSA-status, census region, solo versus group practice, and measures of each provider’s average patient age, sex, race, percent of patients referred to another physician, percent of patients where the physician is the primary care provider, and percent Medicare patients. Included in all model specifications are year and specialty
fixed effects. This creates a difference-in-difference model over physician specialties and time where specialties are differentially affected by reimbursement rate changes since the exposure to relative Medicare prices varies by specialty and time, due to differences in the types of services that are billed to Medicare by each specialty.
Identification of β requires exogenous changes in Medicare payments that are uncorre- lated with other factors which might drive physician-hospital integration. This implies that relatively sharp changes in the reimbursement differential are necessary, otherwise we may be concerned that long-run trends in procedure costs may be correlated with integration of hospitals and physicians. Ody and Dranove (2016) point out that changes in Medicare’s ad- ministrative prices are usually not exogenous, since they are meant to compensate providers for the costs of services and procedures and as such are influenced by factors which affect the cost of running a physician practice. To overcome this potential endogeneity, Ody and
Dranove (2016) use a 2010 change in the method used to calculate Medicare reimbursement rates. This survey change is implicit in my identification strategy, although it is specified differently. CMS also adjusted their methodology for calculating practice expense costs in
2007, with changes phased in from 2007 through 2010.32
32see https://www.cms.gov/Regulations-and-Guidance/Guidance/Transmittals/downloads/R258OTN.pdf
24 It is also likely that the observed rapid changes in relative reimbursement rates for integrated versus independent physicians do not reflect changes in actual procedure costs across the two settings. This is especially true for physician practices which integrate with hospitals and bill as a hospital outpatient department, but still serve a similar patient population, in the same location, and perform similar services.
Integrating vertically is only one potential outcome decision that may be made by physi- cians and hospitals in response to changes in reimbursement. If physicians increase the quantity of services billed (demand inducement), shift billing to other types of services, or move to larger group practices in the face of falling reimbursement rates, then we should not see any effect of relative reimbursement on hospital ownership.
1.6 Empirical Results
Results from estimating equation 1.5 are presented in Table 1.4.33,34 Column 1 presents a basic model with no additional controls besides specialty and year fixed effects.35 Column
2 includes controls for physician practice and patient characteristics. Column 3 includes ad-
33All regression results are estimated with models which utilize the provided NAMCS physician-level survey weights. This is to maintain consistency with reported statistics which utilize the survey weights. If physician-level survey weights are not utilized, the standard errors increase slightly, but magnitudes and interpretation remain largely unchanged. The sample is limited to one observation per physician when survey weights are not used. See appendix for unweighted results. 34I omit psychiatry specialties from all regression models. This is done for two reasons: First, psychiatry services may be reimbursed under the HOPPS in some other care settings including community mental health centers and for certain services such as partial hospitalizations. Psychiatry has the highest average rate of “other” ownership across the sample period. It is not clear that hospital ownership of psychiatry providers will respond in a similar way as other Medical Care specialties. Second, in constructing the relative price measure for psychiatry, there are multiple years for which a sizable amount of services are not matched to Medicare payment rates, possibly resulting in a poor characterization of the exposure of psychiatry to Medicare’s provider-based billing policy. Appendix Tables A.1 - A.3 present models with and without psychiatry included. 35All standard errors are heteroscedastic robust. Clustering standard errors at the physician specialty level does not greatly affect significance and does not alter the interpretation of results. Clustering in the presence of a small number of groups may result in incorrect standard errors (Cameron and Miller, 2015). This is discussed in greater detail in the appendix, but adjusting standard errors for a small number of clusters does not result in substantively different standard errors or significance.
25 ditional Primary Care, Surgical Care, and Medical Care specialty fixed effects.36 Columns
4 and 5 interact the reimbursement ratio measure with aggregate specialty categories.37
Across all specialties combined, columns 1 through 3 show that there is no consistent statis- tically significant effect of changes in relative Medicare reimbursement on hospital ownership of physician practices. There is a positive and significant (10 percent level) response in own- ership only in column 1, although estimated magnitudes are very similar once other controls are included. Interpreting column 1, a 10 percent increase in the relative price paid to inte- grated physicians leads to a 0.8 percentage point increase in hospital ownership. Columns
4 and 5 show that when the responsiveness to relative price is allowed to vary by specialty category, there is evidence that physicians respond differently across specialties. Primary care specialties are the least responsive to changes in relative reimbursement rates; esti- mated coefficients are negative, but insignificant. Estimates for Surgical Care specialties are positive, but insignificant. The largest response in hospital ownership to changes in
Medicare reimbursement is for Medical Care specialties. As the relative price increases by
10 percent, hospital ownership increases by 1.7 to 1.9 percentage points.
To put the results into context, Figure 1.13 uses regression models reported in Table 1.4 to construct two alternate predictions.38 In the first model, hospital ownership is predicted using fitted values from regression models using observed values of covariates. The second prediction constrains Medicare relative prices to remain at their 2005 levels. This is a simple way to compare observed integration versus predicted levels of integration had Medicare relative prices not diverged over time. Results from the model imply that if Medicare rates remained at their 2005 levels, then observed integration for Medical Care specialties would have been about 4 percentage points, or about 32 percent lower in 2015. Surgical care spe-
36Models with and without Primary Care, Surgical Care, and Medical Care specialty category fixed effects are presented since the more detailed specialty indicators are somewhat collinear with the aggregate specialty category indicators, with the exception of some overlap of detailed physician specialties. Primary care specialties include general and family practices, internal medicine, pediatrics, and obstetrics/gynecology. Surgical specialties include general surgery, orthopedic surgery, urology, opthalmology, otolaryngology, and a small number of ob/gyn’s. Medical specialties include cardiovascular disease, dermatology, psychiatry, and neurology, as well as a some pediatrics and fewer Ob/Gyn and general/family practice physicians. 37Instead of omitting one category and interpreting coefficient estimates relative to the omitted category, I include all three categories so that coefficients may be interpreted relative to 0. This makes calculation of the total effect simpler and is equivalent since the interactions include a discrete categorical variable. 38Estimated coefficients from Table 1.4, column 5 are used in predictions.
26 cialties show a smaller gap of about 2 percentage points lower in 2015. Predicted ownership for Primary Care specialties is nearly identical using observed versus 2005 Medicare relative prices.
The negative, although insignificant, estimated coefficient on Medicare reimbursement rates for Primary Care providers (comprised mostly of general and family practice and inter- nal medicine specialties) seems inconsistent with the anticipated effect where it is expected that increases in the relative payment for services furnished in a hospital-owned practice should result in increased hospital ownership. However, even in the raw data, hospital own- ership is increasing for Primary Care providers, and the relative price is flat or declining over most of the sample (see Figures 1.7 and 1.10). There are a couple of plausible expla- nations for this finding: It is possible that hospitals value Primary Care specialties, not for the value of increased reimbursement for a given service, but for the referrals generated by providers in these specialties. The coefficient estimate on relative Medicare prices may also be attenuated if the relative price measure is not accurately capturing the true exposure of a specialty to differential reimbursement rates.
Primary care physicians act as “gatekeepers” who direct patients to other specialists, if necessary. Consistent with this, the percent of patients referred to other physicians is higher for Primary Care specialties, compared to medical and Surgical Care specialties.
Physicians who reported hospital ownership also tend to generate more referrals, however,
Since the NAMCS data is a repeated cross-section, I am unable to tell whether physicians who generate more referrals are more likely to be acquired, or if physician referrals increase only after a hospital acquires the practice. Baker, Bundorf, and Kessler (2016) find that for Medicare patients hospital ownership of a patient’s physician increases the probability that the patient will choose the owning hospital for subsequent inpatient services. Further, hospital ownership results in patients choosing higher cost, lower quality, and hospitals further away from their residence. This behavior suggests that physicians employed by a hospital do tend to send patients to the owning hospital, regardless of how it compares with its competitors.
There may also be measurement error present in the aggregate relative price measure.
Capps, Dranove, and Ody (2017) note that for their particular insurance data provider,
27 the insurance company does not pay a differential for “evaluation and management” visits, which make up a dominant proportion of the services billed for by primary care physicians.
If it is common among private insurers to deviate from Medicare provider-based billing rates for evaluation and management services, then the aggregate, specialty-level relative price exposure measure will be a poor characterization of the actual exposure faced by
Primary Care providers. Primary Care specialties also include pediatricians and obstetrics and gynecology providers. Services provided to Medicare patients, which include mostly over-65 and a small amount of disabled individuals, by providers in these specialties may be very different than the types of services utilized by privately insured patients. These two factors may result in substantial measurement error in the constructed relative price measure for Primary Care physicians, which would bias the estimated coefficient. I am able to test the sensitivity of results with pediatrics and obstetrics and gynecology omitted from regression models (see Appendix Table A.4). Although the estimated coefficient for relative
Medicare price is small and positive for Primary Care physicians, it remains insignificant.
Similarly for Surgical Care specialties, some of the most common procedures are not performed across both settings of care and so the relative price measure may be missing some services which make up a potentially large share of physicians’ revenue and will be measured with more error for these specialties.
1.7 Robustness
1.7.1 Parallel Trends
One of the primary identifying assumptions in any difference-in-differences analysis is that “but for” the treatment, the observed outcome for the treated group would have been similar to the control group. The framework presented in this paper employs a contin- uous treatment measure reflecting exposure of a physician specialty to relative Medicare reimbursement for integrated versus non-integrated providers. As such, there is no clear
“treatment” and “control” group distinction. There is also no distinct pre-treatment and post-treatment period, although much of the change in reimbursement rates results from changes in the methodology CMS uses to update reimbursement rates occurring in 2007 and
28 2010. Examining trends in hospital ownership before these changes occurred for specialties that are eventually exposed to a larger gap in relative prices versus specialties with little change in exposure may still provide some evidence on the reasonableness of the parallel trends assumption.
Figure 1.15 shows trends in hospital ownership for selected physician specialties from
1997 through 2015.39 Cardiology and urology witness the largest changes in relative re- imbursement from 2005 through 2015 (increases of about 47 and 35 percent, respectively), while dermatology and general and family practice witness little change (see Table 1.1).
Average hospital ownership rates remain relatively flat across all selected specialties prior to 2007-2008. After 2007-2008, trends in hospital ownership begin to diverge. Ownership for both cardiology and urology begin to increase, while ownership for general and family practice and dermatology remain relatively flat. A lack of diverging trends in ownership across specialties with widely varying eventual treatment exposure provides some evidence in support of the parallel trends assumption in this analysis.
1.7.2 Medicare Patient Share
Alpert, Hsi, and Jacobson (2017) use geographic, county-level variation in the popula- tion’s share of Medicare-eligible to identify the effect of Medicare Part B payment reforms on vertical integration in oncology. The identifying assumption is that physicians who prac- tice in counties with a higher share of Medicare-eligible individuals (people over age 65) see more Medicare patients as a share of their total patients and have a greater exposure to the Medicare payment reforms. Unfortunately, the public-use NAMCS does not include geographic identifiers, and so I cannot leverage similar geographic variation in patient mix.
The NAMCS data, however, does include an indicator for each patient-physician visit iden- tifying Medicare patients; I use this to calculate an average Medicare patient share for each physician. Following the general methodology of Alpert, Hsi, and Jacobson (2017), I interact physicians’ shares of Medicare patients with the Medicare relative reimbursement
39Years are collapsed into two-year bins for readability. Statistics in Figure 1.15 do not use physician sur- vey weights since these weights are available in the public-use file only after 2004. Weighted and unweighted ownership statistics are generally similar.
29 measure to see if physicians whose practice depends relatively more on Medicare patients are more responsive to changes in Medicare payment rates. Table 1.5 shows model speci-
fications which include an interaction term for physicians’ Medicare patient shares, which is standardized to be mean 0 and have a standard deviation of 1, and Medicare relative reimbursement rates. Results are largely insignificant, although magnitudes are similar to baseline results in Table 1.4. Medical Care specialties still show a significant response to changes in relative Medicare rates, but the only significant Medicare patient-share inter- action is negative for Surgical Care specialties. However, the presented coefficients show only an average across the range of values for both the relative Medicare rates and Medi- care patient share. Figure 1.11 shows various marginal effects of relative Medicare rates over different values of the (standardized) share of Medicare patients. Estimated results in Figure 1.11 are from Table 1.5 column 3. The average Medicare patient share is about
25 percent with a standard deviation just under 24 percent. Figure 1.11 shows that from a standardized Medicare patient share of about .75 through 2.75 standard deviations, the effect of relative Medicare rates is positive and significant at the 10 percent level. Over this range, the average marginal effect of Medicare relative reimbursement ranges from about
0.85 ppt to 1.12 ppt, and there is a slight upward trajectory indicating that physicians who see more Medicare patients are more responsive to changes in Medicare rates. Note that these results are an average across all specialties, allowing the relative effect of Medicare payment gaps between hospital and office settings to vary only with the share of Medicare patients that a physician sees.
Since the NAMCS is a repeated cross-section, I cannot observe the same physician under different ownership or follow his or her patient mix over time. The above analysis compares physicians who see few versus many Medicare patients and necessarily assumes that patient mix, or at least the share of Medicare patients that a physician sees, is exogenous with ownership status. To relax this assumption, I calculate an average of the provider-level
Medicare patient shares for each physician specialty. While there is significant variation in the Medicare patient shares among providers within each specialty, there are also clear
30 differences across specialties.40 Perhaps unsurprisingly, pediatrics and Ob/Gyn’s treat rel- atively few Medicare patients, and cardiologists and urologists see relatively more. The specialty-average Medicare patient share should capture the average exposure of providers within a specialty to Medicare payment changes, but should not be influenced by any particular provider or be systematically related to ownership changes. Table 1.6 presents results using the specialty-level average Medicare patient share interactions. Columns 1 through 3 show that the Medicare relative reimbursement measure alone does not seem to matter, but when interacted with the average Medicare patient share, there is a significant and quite large effect on hospital ownership. Figure 1.12 presents the marginal effects of a change in relative Medicare reimbursement across different values of Medicare patient shares. Columns 4 and 5 of table 1.6 show that this interaction effect is driven by Primary
Care specialties.
The analysis presented in this section provides some evidence that physicians who see a larger share of Medicare patients are more responsive to Medicare reimbursement rate changes. This is expected if providers are truly responding to Medicare fee schedule changes which alter their potential compensation in hospital and office settings. Although reim- bursement for outpatient health care services paid by private insurers do follow Medicare’s provider-based differentials (Capps, Dranove, and Ody, 2017; Koch, Wendling, and Wilson,
2017a), these differentials may not be paid for all services, and depending on the negotiation process, may not be as large as those paid by Medicare (Capps, Dranove, and Ody, 2017).
If this is true, then providers who see relatively more Medicare patients should be more responsive to changes in provider-based payment differentials.
1.7.3 Dynamic Effects
Tables 1.7 and 1.8 introduce lags and leads of the relative price measure to investigate the dynamic effects of Medicare reimbursement rates. Since the integration decision of a hospital and physician is not easily reversible, there may be a lag between the time CMS updates payment rates and a behavioral response in integration. Physicians and hospitals may also
40See appendix Figure A.3.
31 take time to update their beliefs about future reimbursement updates, given a change in observed reimbursement levels. Therefore, lagged values of reimbursement may be relevant for contemporaneous firm ownership decisions, and leads of the reimbursement rate should arguably not drive the integration decision of hospitals and physicians. According to Table
1.7, there is a larger and significant increase in hospital ownership of physician practices in response to lagged (one year) relative reimbursement (columns 1 through 3). This effect is driven by Medical Care and Surgical Care specialties (see column 5), with Primary Care physicians still exhibiting no statistically significant response.41 Column 6 shows that lead values of reimbursement do not have a significant effect on hospital ownership, and the magnitudes are typically closer to zero than contemporaneous and lagged values. Table 1.8 presents a regression model with leads and lags of two years. Overall, results are similar to Table 1.7 where estimated magnitudes for contemporaneous and lag values are stronger predictors of hospital ownership, with lead values typically being insignificant and smaller in magnitude. Medical care specialties do show a positive and statistically significant (10 percent level) response to leads of two years, however this may be due to relatively small year-to-year changes in reimbursement and collinearity between leads and contemporaneous reimbursement. Including two-year leads and lagged values of reimbursement also begins to reduce the sample size and range of the sample period considerably.
Overall Tables 1.7 and 1.8 provide some evidence that lagged values of relative reim- bursement are driving integration between hospitals and physicians. Lead values of reim- bursement have less predictive power for hospital ownership, indicating that the observed response is not due to some spurious correlation driving both reimbursement and ownership.
Figure 1.14 performs a similar exercise as 1.13 and predicts hospital ownership under two different relative price assumptions, using regression specifications from Table 1.7 column
5. In Figure 1.14, both Medical care and Surgical care specialties show gaps in predicted hospital ownership when Medicare relative prices are pinned to their 2005-levels versus al- lowed to change as observed in the data. About one-third of integration for these specialties
41Lags and leads of the relative reimbursement rate are entered into regression models separately since year-to-year changes are highly collinear. Results are similar when contemporaneous, lead, and lag values of reimbursement are included within the same model
32 is due to increases in the relative price paid to providers. Again, there is little difference for Primary care specialties.
1.8 Discussion
1.8.1 Geographic Variation in Hospital Ownership
This paper uses variation in national Medicare fee schedule prices and services provided across specialties to examine the contribution of provider-based billing on hospital ownership of physician practices. This methodology is partly by necessity since the public-use NAMCS data do not have geographic identifiers beyond Census region for most years during the
2005 through 2015 sample period. Therefore, the estimated effect of Medicare’s provider- based billing policy on vertical integration between physicians and hospitals represents an average across many different markets with potentially different landscapes of provider competition, insurer bargaining power, patient types, and other factors. As Brunt and
Bowblis (2014) and McCarthy and Huang (2016) show, local insurer market competition may be a relevant determining factor in the ownership structure of hospital and physician markets. Examining the effect of vertical integration on prices, Capps, Dranove, and Ody
(2017) find that physician offices acquired by hospitals with a larger inpatient market share have subsequently larger price increases than those acquired by a smaller hospital system. It is possible that local provider competition attenuates or exacerbates the response in vertical integration between hospitals and physicians to Medicare price differentials.
Although the (public-use) NAMCS data employed in the main analysis of this paper cannot be used to explore geographic variation in hospital ownership of physician practices, the Provider Utilization and Payment from CMS includes provider-level Medicare utilization data and provides an alternative way to measure integration.42 I calculate the proportion of services and physician payments made in the facility (hospital) setting and office setting. If a physician practice is owned by a hospital and bills as such, then the Medicare claim should
42CMS Provider Utilization and Payment Data is available for Physicians and Other Suppliers for years 2012 through 2015. It includes details on the number of services and dollar amount reimbursed for specific Medicare Part B outpatient services, identified by HCPCS codes, delivered to Medicare patients by a provider by year and setting (office or facility).
33 indicate that the service took place in a facility setting.43 Figure 1.16 shows the proportion of services that take place in a facility versus office setting.44 This measure captures the overall volume of services performed in a hospital versus office for each Hospital Referral
Region. Alternatively, Figure 1.17 shows the proportion of providers in each hospital referral region who perform most (> 95%) of their services in a facility setting. This latter measure captures the proportion of providers who provide services almost exclusively in a facility setting and are most likely directly employed by a hospital.
Both measures show that there is a large geographic variation in where services are delivered. The percent of services delivered in a hospital by hospital referral region varies from a low of about 4 percent to a high of almost 76 percent. The proportion of physician providers who perform greater than 95 percent of services in a hospital varies from a low of 6 percent to a high of 60 percent. The upper-Northeast, Northern Midwest, and parts of the Pacific Northwest tend to have a larger proportion of services delivered in a hospital setting.45 There are numerous potential explanations for this observed variation. According to a GAO report on health insurer concentration, as of 2013 a number of states across the northern U.S., including Idaho, Iowa, Maine, Minnesota, Montana, North Dakota, and
South Dakota, had insurance markets (individual, small group, or large group) where the top three insurers represented over 90 percent of enrollment (GAO, 2014). These states also represent many of the Hospital Referral Regions with the highest proportion of integration, as measured by facility share. However, other areas, such as the Southeast, also include a
43Neprash, Chernew, Hicks, et al. (2015) use a similar methodology for constructing measures of hospital and physician integration. Henry et al. (2018) use the office and facility indicators in the CMS Provider Utilization and Payment data to determine the proportion of services and billing occurring at hospitals, offices, and ambulatory surgery centers. Although measures of facility-share capture the proportion of services performed in a hospital, they do no inform us on the exact employment arrangement. A service may be performed in a facility if the physician has privileges at a hospital, works directly for the hospital either as a hospitalist or laborist or in an outpatient department, or if he or she works in a practice owned by a hospital and the practice is billing under provider-based billing rules. If a practice is hospital owned, but is not billing Medicare as such, then the service claim should reflect an office setting. However, I still use facility-share as an approximate measure of hospital and physician integration. 44I limit the sample of services to those are performed in both facility and office settings at least some of the time; services performed in an office setting either less than 1% of the time or greater than 99% of the time are excluded. I also limit the sample of physician specialties to those included in the NAMCS data. 45This geographic relationship also holds for most of the individual physician specialties that are aggre- gated together in Figures 1.16 and 1.17, indicating that the variation in facility share is not due to different concentrations of specialists across geographies. Figure 1.17 also appears very similar if a 70 or 80 percent facility share cutoff is used instead of 95 percent.
34 group of states with very concentrated health insurance markets, yet do not exhibit higher levels of integration, as measured in Figures 1.16 and 1.17.
1.9 Conclusion
Integration between physicians and hospitals is becoming increasingly common, yet nei- ther the mechanisms driving the formation of this type of organizational structure nor the consequences in terms of service utilization, prices, or quality are completely understood.
This paper finds evidence that Medicare’s provider-based payment policies are in part re- sponsible for the rapid increase in physician-hospital integration among certain physician specialties. Medical Care and, to a lesser extent, Surgical Care specialties respond to Medi- care payment rules which reimburse hospital-owned physician practices at higher rates than independent physician-owned practices. A 10 percent increase in the relative reimbursement rate increases the probability of hospital ownership of Medical Care specialties by about
1.4 to 1.9 percent. In some model specifications, hospital ownership of surgical specialties increase by about 2.1 percent in response to a 10 percent increase in relative reimbursement.
This response explains about one-third of the observed physician-hospital integration from
2005 to 2015. Primary Care physicians do not seem as responsive to changes in the rela- tive Medicare reimbursement rate paid to hospital-owned practices versus physician-owned.
There is some evidence that physicians whose practice depends more on Medicare patients are more responsive to changes in Medicare’s administrative prices. This helps explain some of the lack of response for Primary Care specialties.
This is an important finding for two reasons. First, finding a positive effect of disparities in Medicare reimbursement rates paid to hospital-owned versus independent physician prac- tices should not be a forgone conclusion where “incentives matter.” Physicians who give up owning and operating their own practice as an entrepreneur in favor of hospital ownership or employment are making a potentially career-altering, irreversible decision. Policy-makers should carefully consider what effect payment policies will have on hospital and physician market structures. Second, if hospital acquisitions of physician practices are undertaken only to take advantage of Medicare reimbursement policies and mechanically increase hos-
35 pitals’ profits from providing health care services in these facilities, with no substantial change in how or where services are delivered, then we should be skeptical that integration will lead to lower costs or better quality of care. This paper does not estimate any impact on health care quality or cost, nor does it calculate benefits or costs to patients. However, cost savings from eliminating or narrowing this differential can be substantial. MedPAC reports estimate a cost savings of $900 million per year by reducing fee differentials for only a group of selected services (MedPAC, 2013). A 2014 Office of Inspector General report estimated that CMS could save $15 billion over the 2012 through 2017 period by equaliz- ing hospital outpatient department and (typically lower) ambulatory surgery center rates
(OIG, 2014). Payment and utilization data from CMS show that for 2012 to 2015, Part B services provided in a hospital outpatient department setting made up around 17 percent of total utilization but 40 percent of Medicare payments (or almost $66 billion).46 Further, providers who performed at least 95 percent of procedures in a hospital outpatient setting incurred an additional $7,300 to $11,200 in annual Medicare payments, versus providers who perform relatively more procedure in an office setting.47 This implies roughly $340 to $520 billion in annual Medicare payments made to physicians who work primarily in a hospital and generate additional Medicare reimbursement through facility fees. Given the quick pace of health care expenditure growth in the United States, the large share of spending that physician and hospital services comprise, and the general trend towards more consolidated health care markets with less competition, this is an area that should continue to be examined closely in the future.
This paper adds to the growing literature addressing vertical integration between hos- pitals and physicians. There has been little recent work examining the motivation for
46Source: CMS Provider Payment and Utilization public-use files from 2012 through 2015. Specialties were limited to those included in the NAMCS data, services were excluded if they were either always or never performed in both office and hospital settings. Hospital outpatient department setting is inferred from the “facility" place code and hospital payments are estimated by adding in the OPPS payment rate, similar to Henry et al. (2018). 47A simple regression analysis using the CMS Provider Payment and Utilization data from 2012 to 2015 shows that Medicare reimbursements, including OPPS payments, are roughly $7,300 to $11,200 higher if the physician performs at least 95 percent of his or her services in a “facility" setting. Regression controls include service volume, specialty and year indicators, and alternatively physician fixed-effects. Similar results are obtained using 80, 85, and 90 percent facility-share cutoffs.
36 physician-hospital integration. The only studies which measure the impact of Medicare payment rates on vertical integration find somewhat conflicting evidence (Alpert, Hsi, and
Jacobson, 2017; Ody and Dranove, 2016). Given the heterogeneous effects of integration on prices, we may not expect that all physician specialties should respond in the same way to changes in Medicare payment rates (Capps, Dranove, and Ody, 2017). This paper is the first to show that integration between physicians and hospitals in response to Medicare payment rules may vary across physician specialties.
CMS has recently addressed the disparities in provider-based payments and effective
January 2017, a hospital acquired (or built) practice that is located “off-campus” may bill services under the HOPPS, thus eliminating the site-based payment methodology. However, most hospital-owned practices are grandfathered in and may continue to bill services at a higher rate.48 Future research may examine whether site-neutral payments slow the growth of hospital-owned practices. The findings in this paper suggest that for some specialties, such as Primary Care physicians, differential Medicare rates may be less relevant for the integration decision, and that other factors will potentially continue to drive physicians to hospital employment. These include risk-based capitated payments, required investment in electronic health records systems, pushes for more coordinated care and less fragmentation among health care providers, and increasingly complex regulations.
48Section 603 of the Bipartisan Budget Act of 2015 requires “site-neutral” payments for off- campus provider-based departments. Any provider-based department operating, or under certain cir- cumstances planned, before November 2, 2015 may continue billing under the HOPPS. This policy went into effect January 1, 2017. https://www.cms.gov/Newsroom/MediaReleaseDatabase/Fact-sheets/ 2016-Fact-sheets-items/2016-11-01-3.html
37 Table 1.1: Medicare Reimbursement by Specialty
2005 2015 PO HOPD Ratio PO HOPD Ratio % Change Cardiology 240.12 252.88 1.25 665.43 938.89 1.84 47.34 Dermatology 195.04 302.69 1.73 248.73 430.67 1.85 6.93 General/family practice 79.89 122.56 1.65 111.48 184.46 1.76 6.97 General surgery 137.96 208.86 1.58 719.38 1050.83 1.87 18.30 Internal medicine 94.39 133.91 1.59 127.98 212.59 1.72 8.22 Neurology 160.90 182.93 1.33 192.76 266.68 1.52 13.83 Ob/Gyn 102.11 145.35 1.51 119.29 230.45 1.83 20.67 Ophthalmology 167.92 249.45 1.54 146.83 280.42 1.93 24.86 Orthopedic surgery 111.52 165.61 1.72 162.82 264.29 2.24 29.99 Otolaryngology 121.08 210.97 1.62 246.79 449.68 1.92 18.20 Pediatrics 93.55 130.95 1.56 126.68 207.38 1.77 13.67 Psychiatry 89.15 163.76 1.94 95.14 158.52 1.72 -11.18 Urology 240.55 338.25 1.73 158.70 408.69 2.33 34.88 Source: Medicare Physician Fee Schedule, Hospital Outpatient Prospective Payment Sys- tem. Service revenue is derived from CMS Utilization Data for 2003 and 2007 through 2015. Physician office (PO) and hospital outpatient department (HOPD) rates are expressed in dollars. PO, HOPD, and the ratio of HOPD to PO rates are indexed values constructed by weighting the reimbursement rate for a service (defined by a HCPCS code) by the total physician office revenue share of that service for providers in that specialty. The sample of services is limited to those with between 1% and 99% of revenue geneerated in a facility setting.
38 Table 1.2: 2005 - 2015 NAMCS: Mean Statistics by Practice Ownership
Physician Owned Hospital Owned Other Owned Medicare HOPD/PO 1.70 1.70 1.70 (0.00) (0.01) (0.00) Located in MSA 90.13 82.44 88.71 (0.36) (1.44) (1.03) Solo practice 41.37 12.71 9.48 (0.62) (1.45) (0.90) % Medicare visits 24.09 23.00 20.29 (0.29) (0.86) (0.72) % Private insured visits 52.94 48.74 49.35 (0.36) (1.14) (1.08) Avg. patient age 46.33 43.04 43.24 (0.23) (0.82) (0.63) % patients female 58.57 59.43 59.00 (0.27) (0.84) (0.74) % patients black 10.23 11.19 10.53 (0.23) (0.78) (0.61) % patients primary care 46.71 52.45 53.15 (0.59) (1.84) (1.46) % patients referred to other phys. 7.47 9.97 11.12 (0.16) (0.58) (0.48) N 11,319 1,120 1,624 Source: NAMCS Public-Use file 2005 - 2015. Standard errors are in parenthesis. Statistics are weighted using physician-level weights. Physician owned indicates physician or physician group ownership, hospital owned indicates medical or academic health center ownership or other hospital ownership. Other ownership includes insurance company, health plan, HMO, or other health care corporation ownership.
39 Table 1.3: 2005 - 2015: Summary Statistics by Specialty
General Internal Ob/Gyn Pediatrics General Ophthalmology Orthopedic family practice medicine surgery surgery % Physician owned 73.1 80.8 82.3 79.2 81.2 92.8 88.6 % Hospital owned 10.1 6.2 7.1 8.8 8.9 2.5 4.8 % Other owned 16.7 13.0 10.6 12.1 9.8 4.7 6.6 Medicare HOPD/PO 1.7 1.6 1.7 1.6 1.8 1.6 2.0 Located in MSA 81.5 90.9 92.5 91.6 82.9 91.9 88.5 Solo practice 35.1 40.0 32.0 24.1 36.0 36.9 26.3 % Medicare visits 22.3 34.5 6.2 1.4 26.8 46.4 25.5 % Private insured visits 52.7 47.3 68.8 62.2 53.4 38.7 52.1 Avg. patient age 48.0 57.8 37.8 7.6 53.6 61.4 51.9 % patients female 57.7 56.7 99.6 48.2 59.8 58.7 53.5 % patients black 9.5 12.6 14.0 11.5 9.9 10.7 8.7
40 % patients p.c.p. 86.2 86.1 15.8 86.6 3.6 2.0 2.2 % patients referred to other phys. 11.1 13.7 4.5 6.0 8.1 4.1 5.7 N 2,572 1,318 1,082 1,567 854 905 980 Otolaryngology Urology Cardiology Dermatology Neurology Psychiatry % Physician owned 88.2 88.6 82.0 90.0 84.0 77.2 % Hospital owned 5.5 5.9 9.1 2.9 7.7 5.3 % Other owned 6.2 5.4 8.9 7.2 8.4 17.6 Medicare HOPD/PO 1.8 2.0 1.6 1.7 1.5 1.9 Located in MSA 91.8 92.0 94.8 95.7 94.8 95.2 Solo practice 33.5 27.0 24.3 44.6 36.8 66.1 % Medicare visits 22.0 41.5 51.4 28.7 30.6 12.6 % Private insured visits 60.5 44.7 36.4 59.5 46.9 41.0 Avg. patient age 43.9 60.5 66.4 52.9 51.4 41.6 % patients female 53.2 26.2 48.1 55.1 57.5 56.3 % patients black 7.1 8.0 9.9 4.1 8.7 7.3 % patients p.c.p. 2.0 2.7 11.9 2.1 2.7 4.8 % patients referred to other phys. 6.6 4.2 7.3 1.9 10.6 2.9 N 677 746 906 665 798 993 Source: NAMCS Public-Use file 2005 - 2015. Statistics are weighted using physician-level weights. Physician owned indicates physician or physician group ownership, hospital owned indicates medical or academic health center ownership or other hospital ownership. Other ownership includes insurance company, health plan, HMO, or other health care corporation ownership. Table 1.4: Hospital Ownership of Physician Practices
(1) (2) (3) (4) (5) ln(HOPD/PO) 0.88* 0.84 0.85 (0.54) (0.52) (0.52) Primary care × ln(HOPD/PO) -0.93 -1.00 (0.75) (1.01) Surgical care × ln(HOPD/PO) 0.86 0.64 (0.73) (0.76) Medical care × ln(HOPD/PO) 1.74*** 1.94*** (0.64) (0.64) Region effects No Yes Yes Yes Yes Other controls No Yes Yes Yes Yes Specialty category No No Yes No Yes N 13,070 13,070 13,070 13,070 13,070 Source: NAMCS Public-Use file 2005 - 2015. * P<0.1, ** P<0.5, *** P<0.01. Robust standard errors are presented in parenthesis. Dependent variable is a binary indicator for hospital ownership. Coefficients have been multiplied by 100 and are interpreted as a percentage point in hospital ownership given a 10 percent increase in HOPD/PO. Year and specialty fixed effects are included in all specifications. Controls include an MSA/Non-MSA indicator, census region, solo versus group practice, percent of visits where the physician is the primary care provider, percent visits where the patient is referred to another physicians, percent Medicare and privately insured patient visits, and and average patient age, sex, and race. Models are weighted using physician-level survey weights. Psychiatry is excluded.
41 Table 1.5: Hospital Ownership of Physician Practices - Medicare Patient Share Interactions
(1) (2) (3) (4) (5) ln(HOPD/PO) 0.78 0.75 0.75 (0.58) (0.57) (0.57) ln(HOPD/PO)×Std. % Medicare patients 0.21 0.12 0.14 (0.27) (0.27) (0.27) Primary care × ln(HOPD/PO) -0.43 -0.67 (1.12) (1.09) Surgical care × ln(HOPD/PO) 0.79 0.94 (0.83) (0.81) Medical care × ln(HOPD/PO) 1.82* 1.66* (1.01) (0.97) Primary care×ln(HOPD/PO)×Std. % Medicare patients 1.01 1.03 (0.92) (0.92) Surgical care×ln(HOPD/PO)×Std. % Medicare patients -0.58 -0.73* (0.39) (0.39) Medical care×ln(HOPD/PO)×Std. % Medicare patients 0.39 0.43 (0.61) (0.59) Region effects No Yes Yes No Yes Other controls No Yes Yes No Yes Specialty category No No Yes Yes Yes N 13,070 13,070 13,070 13,070 13,070
Source: NAMCS Public-Use file 2005 - 2015. * P<0.1, ** P<0.5, *** P<0.01. Robust standard errors are presented in parenthesis. Dependent variable is a binary indicator for hospital ownership. Coefficients have been multiplied by 100 and are interpreted as a percentage point in hospital ownership given a 10 percent increase in HOPD/PO. Year and specialty fixed effects are included in all specifications. Controls include an MSA/Non-MSA indicator, census region, solo versus group practice, percent of visits where the physician is the primary care provider, percent visits where the patient is referred to another physicians, percent Medicare and privately insured patient visits, and and average patient age, sex, and race. Models are weighted using physician-level survey weights. Psychiatry is excluded.
42 Table 1.6: Hospital Ownership of Physician Practices - Medicare Patient Share Interactions
(1) (2) (3) (4) (5) ln(HOPD/PO) -0.07 -0.17 -0.18 (0.66) (0.64) (0.64) ln(HOPD/PO)×Avg. % Medicare patients 1.08*** 1.10*** 1.13*** (0.38) (0.37) (0.37) Primary care × ln(HOPD/PO) 0.19 -0.03 (1.18) (1.15) Surgical care × ln(HOPD/PO) 0.82 1.10 (0.91) (0.91) Medical care × ln(HOPD/PO) 1.04 0.75 (2.49) (2.41) Primary care×ln(HOPD/PO)×Avg. % Medicare patients 1.77** 1.84** (0.85) (0.83) Surgical care×ln(HOPD/PO)×Avg. % Medicare patients -0.66 -0.99 (0.77) (0.76) Medical care×ln(HOPD/PO)×Avg. % Medicare patients 0.58 0.69 (1.42) (1.37) Region effects No Yes Yes No Yes Other controls No Yes Yes No Yes Specialty category No No Yes Yes Yes N 13,070 13,070 13,070 13,070 13,070 Source: NAMCS Public-Use file 2005 - 2015. * P<0.1, ** P<0.5, *** P<0.01. Robust standard errors are presented in parenthesis. Dependent variable is a binary indicator for hospital ownership. Coefficients have been multiplied by 100 and are interpreted as a percentage point in hospital ownership given a 10 percent increase in HOPD/PO. Year and specialty fixed effects are included in all specifications. Controls include an MSA/Non-MSA indicator, census region, solo versus group practice, percent of visits where the physician is the primary care provider, percent visits where the patient is referred to another physicians, percent Medicare and privately insured patient visits, and and average patient age, sex, and race. Models are weighted using physician-level survey weights. Psychiatry is excluded.
43 Table 1.7: Hospital Ownership of Physician Practices - Lags and Leads of Relative Price
(1) (2) (3) (4) (5) (6) ln(HOPD/PO) 0.80 (0.56) ln(HOPD/PO t−1) 1.59** (0.65) ln(HOPD/PO t+1) 0.73 (0.64) Primary care × ln(HOPD/PO) -0.35 (1.11) Surgical care × ln(HOPD/PO) 1.14 (0.80) Medical care × ln(HOPD/PO) 1.38* (0.73) Primary care × ln(HOPD/PO t−1) -2.47 (1.61) Surgical care × ln(HOPD/PO t−1) 2.17** (1.01) Medical care × ln(HOPD/PO t−1) 1.44** (0.68) Primary care × ln(HOPD/PO t+1) 0.77 (1.05) Surgical care × ln(HOPD/PO t+1) 0.43 (0.80) Medical care × ln(HOPD/PO t+1) 1.01 (0.91) Region effects Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Specialty category Yes Yes Yes Yes Yes Yes N 11,011 11,011 11,011 11,011 11,011 11,011 Source: NAMCS Public-Use file 2005 - 2015. * P<0.1, ** P<0.5, *** P<0.01. Robust standard errors are presented in parenthesis. Dependent variable is a binary indicator for hospital ownership. Coefficients are interpreted as a percentage point in hospital ownership given a 10 percent increase in HOPD/PO. Year and specialty fixed effects are included in all specifications. Controls include an MSA/Non-MSA indicator, census region, solo versus group practice, percent of visits where the physician is the primary care provider, percent visits where the patient is referred to another physicians, percent Medicare and privately insured patient visits, and and average patient age, sex, and race.
44 Table 1.8: Hospital Ownership of Physician Practices - Two-Year Lags and Leads of Relative Price
(1) (2) (3) (4) (5) (6) ln(HOPD/PO) 1.43 (0.94) ln(HOPD/PO t−2) 2.17*** (0.75) ln(HOPD/PO t+2) 0.97 (0.76) Primary care × ln(HOPD/PO) -0.19 (2.17) Surgical care × ln(HOPD/PO) 1.91 (1.44) Medical care × ln(HOPD/PO) 1.26 (0.88) Primary care × ln(HOPD/PO t−2) 0.21 (1.93) Surgical care × ln(HOPD/PO t−2) 2.39** (1.14) Medical care × ln(HOPD/PO t−2) 2.02** (0.79) Primary care × ln(HOPD/PO t+2) 1.47 (0.95) Surgical care × ln(HOPD/PO t+2) -0.96 (1.52) Medical care × ln(HOPD/PO t+2) 3.45* (2.00) Region effects Yes Yes Yes Yes Yes Yes Other controls Yes Yes Yes Yes Yes Yes Specialty category Yes Yes Yes Yes Yes Yes N 8,603 8,603 8,603 8,603 8,603 8,603 Source: NAMCS Public-Use file 2005 - 2015. * P<0.1, ** P<0.5, *** P<0.01. Robust standard errors are presented in parenthesis. Dependent variable is a binary indicator for hospital ownership. Coefficients are interpreted as a percentage point in hospital ownership given a 10 percent increase in HOPD/PO. Year and specialty fixed effects are included in all specifications. Controls include an MSA/Non-MSA indicator, census region, solo versus group practice, percent of visits where the physician is the primary care provider, percent visits where the patient is referred to another physicians, percent Medicare and privately insured patient visits, and and average patient age, sex, and race.
45 Figure 1.1: Percent of Hospitals Engaged in Affiliations With Physicians
40%
30%
20% Percent of Hospitals
10%
0% 1994 1996 1998 2000 2002 2004 2006 2008 2010 2012
Independent Practice Association Managment Service Organization Group Practice Without Walls Physician Hospital Organization
Source: American Hospital Association 2004 Trendwatch Chartbook Appendix 2, Table 2.5 and 2015 Trendwatch Chartbook Appendix, Table 2.4
46 Figure 1.2: Physicians’ Place of Employment
55%
50%
45%
40% Percent of Physicians
35%
30% 2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
Physician Offices Hospitals
Source: American Community Survey PUMS 1-year estimates. Estimates calculated using person-level weights. Industry codes 7970 (Offices of Physicians), 8080 (Offices of other Medical Professionals), and 8190 (Hospitals). Occupation codes 3060 (Physicians/Surgeons).
47 Figure 1.3: Physicians’ Place of Employment by Physician Age Group
Physician Office 80%
70%
65 and older 60%
50% 45 to 64
40% Percent of Physicians 30% 25 to 44
20%
10%
Hospital 80%
70% 25 to 44
60%
50%
40% 45 to 64
Percent of Physicians 30%
65 and older 20%
10% 2005 2007 2009 2011 2013 2015 Source: American Community Survey PUMS 1-year estimates. Estimates weighted using person-level weights. Industry codes 7970 (Offices of Physicians), 8080 (Offices of other Medical Professionals), and 8190 (Hospitals). Occupation codes 3060 (Physicians/Surgeons). 48 Figure 1.4: Other Medical Occupations’ Place of Employment
80%
70%
60%
50%
40%
30% Percent of Occupation 20%
10%
2005 2006 2007 2008 2009 2010 2011 2012 2013 2014 2015
RN - Physician Offices LPN - Physician Offices PA - Physician Offices RN - Hospitals LPN - Hospitals PA - Hospitals
Source: American Community Survey PUMS 1-year estimates. Estimates calculated using person-level weights. Industry codes 7970 (Offices of Physicians), 8080 (Offices of other Medical Professionals), and 8190 (Hospitals). Occupation codes 3130 (pre-2010 RN's), 3255 (post-2010 RN's), 3256 (Nurse Anesthetists), 3257 (Nurse Midwives), 3258 (Nurse Practitioners), 3500 (LPN's), and 3110 (Physician Assistants).
49 Figure 1.5: Distribution of Facility-Shares for Medicare Services
40%
30%
20%
Percent of HCPCS codes 10%
0% 10% 20% 30% 40% 50% 60% 70% 80% 90% 100% Facility-share
5% - 95% 1% - 99% All
Source: CMS Medicare utilization 2003, 2007-2015.
50 Figure 1.6: Percent of Total Services and Revenue by Facility-Share
100%
80%
60%
40%
20%
0% Percent of HCPCS Percent of Charges
1% to 99% 5% to 95% All HCPCS facility-share facility-share
51 Figure 1.7: Relative Medicare Reimbursement by Specialty
Cardiology Dermatology General and Family Practice General surgery Internal medicine 2.4 2.4 2.4 2.4 2.4
2.2 2.2 2.2 2.2 2.2
2.0 2.0 2.0 2.0 2.0
1.8 1.8 1.8 1.8 1.8
1.6 1.6 1.6 1.6 1.6 HOPD/PO Ratio
1.4 1.4 1.4 1.4 1.4
1.2 1.2 1.2 1.2 1.2 2005 2008 2011 2014 Neurology Obstetrics and gynecology Ophthalmology Orthopedic surgery Otolaryngology 2.4 2.4 2.4 2.4 2.4
2.2 2.2 2.2 2.2 2.2
2.0 2.0 2.0 2.0 2.0 52 1.8 1.8 1.8 1.8 1.8
1.6 1.6 1.6 1.6 1.6
1.4 1.4 1.4 1.4 1.4
1.2 1.2 1.2 1.2 1.2
Pediatric medicine Psychiatry Urology 2.4 2.4 2.4
2.2 2.2 2.2
2.0 2.0 2.0
1.8 1.8 1.8
1.6 1.6 1.6
1.4 1.4 1.4
1.2 1.2 1.2
Weighted average reimbursement rate measures are constructed using 2003 and 2007 through 2015 Utilization Crosswalk Data from CMS. 2003 utilization data is used for year 2005 and 2007 utilization data is used for year 2006. Utilization weights are estimated for each specialty and year and sum to one. The sample is limited to HCPCS service codes with an average of between 1% and 99% of revenue generated in a facility setting. Figure 1.8: Distribution of Relative Medicare Reimbursement by Service and Year
.6
2005 2008 2011 2014
.4 Density
.2
0 0 2 4 6 8 10 12 14 HOPD/PO ratio Source: 2005-2015 CMS Medicare Physician Fee Schedule RVU files and Hospital Outpatient Prospective Payment System Addendum B.
53 Figure 1.9: Physician, Hospital, and Other Ownership of Physician Practices by Specialty
100% 90% 80% 70% 60% 50% 40% 30% Percent of Practices 20% 10% 0%
Urology Neurology PediatricsPsychiatry Dermatology Ophthalmology Otolaryngology Other specialties General surgeryInternal medicine Orthopedic surgery
Cardiovascular diseases Obstetrics and gynecology General & Family practice
Physician Owned Hospital Owned Other
Source: NAMCS Public-Use file 2005-2015. Statistics are calculated using physician-level survey weights. Other ownership includes insurance company, health plan, HMO, or other health care corporation ownership.
54 Figure 1.10: Relative Medicare Reimbursement and Hospital Ownership by Specialty
Cardiology Dermatology General and Family practice General surgery Internal medicine 2.4 30 2.4 30 2.4 30 2.4 30 2.4 30
2.2 25 2.2 25 2.2 25 2.2 25 2.2 25
2.0 20 2.0 20 2.0 20 2.0 20 2.0 20
1.8 15 1.8 15 1.8 15 1.8 15 1.8 15
1.6 10 1.6 10 1.6 10 1.6 10 1.6 10
1.4 5 1.4 5 1.4 5 1.4 5 1.4 5
1.2 0 1.2 0 1.2 0 1.2 0 1.2 0 2005 2007 2009 2011 2013 2015 Neurology Obstetrics and gynecology Ophthalmology Orthopedic surgery Otolaryngology 2.4 30 2.4 30 2.4 30 2.4 30 2.4 30
2.2 25 2.2 25 2.2 25 2.2 25 2.2 25
2.0 20 2.0 20 2.0 20 2.0 20 2.0 20 55 1.8 15 1.8 15 1.8 15 1.8 15 1.8 15
1.6 10 1.6 10 1.6 10 1.6 10 1.6 10
1.4 5 1.4 5 1.4 5 1.4 5 1.4 5
1.2 0 1.2 0 1.2 0 1.2 0 1.2 0
Pediatrics Psychiatry Urology 2.4 30 2.4 30 2.4 30
2.2 25 2.2 25 2.2 25
2.0 20 2.0 20 2.0 20 HOPD/PO Ratio (Left) 1.8 15 1.8 15 1.8 15 % Hospital Ownership (Right) 1.6 10 1.6 10 1.6 10
1.4 5 1.4 5 1.4 5
1.2 0 1.2 0 1.2 0
Weighted average reimbursement rate measures are constructed using 2003 and 2007 through 2015 Utilization Crosswalk Data from CMS. 2003 utilization data is used for year 2005 and 2007 utilization data is used for year 2006. Utilization weights are estimated for each specialty and year and sum to one. The sample is limited to HCPCS service codes with an average of between 1% and 99% of revenue generated in a facility setting. Figure 1.11: Medicare Patient Share Interactions with Relative Medicare Rates
.3
.2
.1
on Hospital Ownership (ppt) 0 Marginal Effect of Relative Price
-.1 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 2.5 3.0 Standardized % Medicare Patients Source: NAMCS Public-Use file 2005-2015. Dotted lines represent 95% confidence intervals.
56 Figure 1.12: Specialty-Level Medicare Patient Share Interactions with Relative Medicare Rates
.4
.2
0
on Hospital Ownership (ppt) -.2 Marginal Effect of Relative Price
-.4 -1.5 -1.0 -0.5 0.0 0.5 1.0 1.5 2.0 Standardized % Medicare Patients Source: NAMCS Public-Use file 2005-2015. Dotted lines represent 95% confidence intervals.
57 Figure 1.13: Predicted Hospital Ownership Rates
Medical Care Surgical Care 14 14
12 12
10 10
8 8
6 6
4 4
2 2 2005 2007 2009 2011 2013 2015 2005 2007 2009 2011 2013 2015 Primary Care 14
12
10 Predicted % Hospital Ownership 8 Predicted % Hospital Ownership at 2005 relative prices 6
4
2 2005 2007 2009 2011 2013 2015 Predicted values are based upon model specifications shown in coulumn 5, of Table 4.
58 Figure 1.14: Predicted Hospital Ownership Rates - Lagged Models
Medical Care Surgical Care
14 14
12 12
10 10
8 8
6 6
4 4
2 2
2005 2007 2009 2011 2013 2015 2005 2007 2009 2011 2013 2015 Primary Care
14
12
10 Predicted % Hospital Ownership 8 Predicted % Hospital Ownership at 2005 relative prices 6
4
2
2005 2007 2009 2011 2013 2015 Predicted values are based upon model specifications shown in column 5, of Table 5.
59 Figure 1.15: “Pretreatment” Trends in Hospital Ownership
30
25
20
15
10 % Hospital Ownership
5
0 1997-1998 2001-2002 2005-2006 2009-2010 2013-2014
Cardiology Urology General and Family Practice Dermatology
Source: NAMCS Public-Use file 1997-2015. Statistics are unweighted since physician survey weights are only availble in the public-use file beginning in 2005.
60 Figure 1.16: Facility-Share of Outpatient Physician Services Delivered to Medicare Patients: 2012 - 2015 61 Figure 1.17: Proportion of Medicare Providers Who Perform At Least 95% of Outpatient Services in a Hospital: 2012 - 2015 62 Chapter 2: Outsourcing Fraud Enforcement – Whistleblower Laws and Medi- caid Expenditure
2.1 Introduction
The United States is a global leader in health care spending, with national health expen- ditures reaching $3.2 trillion in 2015, while spending on Medicaid was $532 billion (NCHS,
2016; Martin et al., 2016). Complex billing requirements combined with vulnerable pa- tient populations and sheer size of the Medicaid program create opportunities for fraud and abuse. Health care fraud comprises up to 10 percent of total health care expenditures and can manifest itself through kickbacks, unnecessary services, over-billing, or illegal marketing and promotion (FBI, 2012; Morris, 2009). Medicare and Medicaid fraud ranges from $30 billion to $98 billion (Berwick and Hackbarth, 2012).
Underlying the problem of health care fraud in public programs is a principal-agent problem; the principal – the government – intends to pay for appropriate goods and ser- vices, but agents have incentives to provide additional services which may go beyond those guidelines, whether for profit motives or disagreements regarding the appropriateness of care. Certain features of the health care system exacerbate the misalignment of incen- tives: reimbursement is mostly fee-for-service where claims are assumed to be legitimate unless there are indications otherwise, claims-processing tends to be highly automated, and verification is deemphasized making monitoring of inappropriate claims extremely difficult
(Sparrow, 2008). Health care providers also bear responsibility for their patients’ well- being, and a fee-for-service system with minimal patient cost-sharing incentives results in potential over-utilization of treatments and tests with marginal positive expected medical benefit (Buchanan, 1988).
Allegations of fraud are pervasive within health care. Hospitals engage in “upcoding” of patients into more highly reimbursed diagnoses groups (Silverman and Skinner, 2004; Dafny,
2005). Physicians similarly upcode patients to take advantage of Medicare fee differentials
(Brunt, 2011) and overbill for services (Fang and Gong, 2017). Pharmaceutical companies report inflated benchmark prices for Medicaid reimbursement purposes (Alpert, Duggan,
63 and Hellerstein, 2013), misclassify drugs to avoid required rebate payments1, deceptively market prescription drugs2, and use inappropriate “detailing” techniques and financial in- centives to promote “off-label” uses to physicians (Stafford, 2008; Kesselheim, Mello, and
Studdert, 2011).
The United States government has, in the last decade, more aggressively pursued fraud and deterrence with False Claims Act (FCA) –“whistleblower” statutes.3 Although litiga- tion for health care fraud under the FCA has increased, little is known about deterring fraudulent behavior, which is a key measure of success.4 A frequent motivation for addi- tional administrative spending is that the deterrent effect is large (Aaron, 2015). However there is little evidence to support such a claim, with one notable exception, Becker, Kessler, and McClellan (2005).
This paper measures the deterrent effect of state FCA statutes and Medicaid Fraud
Control Unit (MFCU) expenditures on Medicaid program spending. The main contribution is quantifying the deterrent effect of whistleblower laws. To do so we extensively document state FCAs and amendments over time. Some surveys have summarized state FCA laws, but none provide sufficient information for quantitative analysis. We focus our attention on
Medicaid program spending where state resources on anti-fraud efforts directly impact state budgets. We examine both Medicaid prescription drug expenditures, where penalties and whistleblower provisions provide strong incentives to deter fraud, and also other services where the incentives are weaker.
Using a difference-in-difference analysis, we find that the presence of such laws has a
1Mylan paid $465 million to the DOJ to resolve questions regarding its classification of EpiPens as generic, rather than brand name for the Medicaid Drug Rebate Program. 2Perdue Pharma and three of its executives plead guilty in 2007 for the misbranding of OxyContin and paid almost $635 million in civil and criminal charges for making claims that the drug was less addictive that its competitors due to its extended release nature. See http://www.nytimes.com/2007/05/10/business/11drug-web.html. 3We use the term “fraud” as consistent with the legal definition. This includes “knowingly submitting false statements or making misrepresentations of fact to obtain a federal health care payment for which no entitlement would otherwise exist, knowingly soliciting, paying, and/or accepting remuneration to induce or reward referrals for items or services reimbursed by federal health care programs, or making prohibited referrals for certain designated health services” (HHS, 2016). 4CMS reports “Success in combating health care fraud, waste, and abuse is measured not only by convictions, but also by effective deterrent efforts.” See https://www.cms.gov/medicare-medicaid- coordination/fraud-prevention/fraudabuseforconsumers/report_fraud_and_suspected_fraud.html
64 sizable effect on prescription drug expenditures. We estimate that per-eligible prescription drug expenditure falls by 21 percent, a sizable cost-saving from the net $22 billion spent per year on Medicaid prescription drugs (MACPAC, 2016).5 There are clear reasons to expect the FCA laws to matter for prescription drugs: highly concentrated, deep-pocketed pharmaceutical companies, a large number of transactions, and a somewhat subjective definition of fraud. At the same time, there is no effect on other service groups which is consistent with anecdotal evidence and actual investigations under the FCA. We find no direct effect of MFCU dollars on Medicaid expenditure.
We further test for the deterrent effect of FCAs through analysis of more fine-grained
Medicaid prescription drug utilization data. Using a triple-difference model, we find that in states with FCAs, after passage of the law, spending on “off-label” prone categories of drugs is more responsive than non-off-label prone categories relative to states without their own FCA. Medicaid prescription drug spending on off-label prone drugs falls by 11 to 14 percent and prescription volume falls by 10 to 14 percent in states with a FCA enacted.
These results are again consistent with FCAs primarily being targeted at pharmaceutical manufacturers for off-label marketing and promotion.
The remainder of this chapter is organized as follows: section 2.2 reviews literature on
FCAs, fraud, and deterrence focusing on private whistleblower litigation, health care fraud and Medicare/Medicaid. Section 2.3 provides institutional background on the False Claims
Act. Section 2.4 discusses how state FCAs may matter differentially for different types of fraud and lays out the conceptual model for our estimation strategy. Section 2.5 describes our data sources. Section 2.6 lays out our empirical model, potential issues, and presents our results. Section 2.7 extends our analysis to specific therapeutic classes of drugs using
Medicaid drug utilization data. Finally, section 2.8 concludes and discusses implications of our findings.
5Medicaid spent approximately $42 billion on prescription drugs in 2014 and collected about $20 billion in rebates.
65 2.2 Literature
2.2.1 Fraud Committed Against the Government
There is a large literature concerning defrauding the government through tax evasion and the role of detection, penalties, and public disclosure which builds upon Allingham and Sandmo (1972). Much like our emphasis on the FCA, some recent work focuses on outsourcing tax enforcement to private parties. For example, Naritomi (2016) examines the deterrence effect of consumer reporting (whistleblowing) of business value-added taxes in
Brazil. Kumler, Verhoogen, and Frías (2015) study third-party employer reporting wages in Mexico. Both show that whistleblower provisions are an effective tool for increasing compliance. Other work focuses on the role of supplementary, third-party information. Ex- amples include Klevin et al. (2011) who analyze a randomized income-tax audit experiment in Denmark, Pomeranz (2015) who studies value added tax enforcement among Chilean
firms, Casaburi and Troiano (2016) who examine property tax evasion in Italy, Carrillo,
Pomeranz, and Singhal (2017) who study corporate income tax reporting in Ecuador, and
Slemrod et al. (2017) who study small-business tax compliance. Work on third-party in- formation finds that tax compliance increases with visibility to the public and enforcing agency, perceived risk of detection, and threat of enforcement.
Private enforcement is also an important source of deterrence and remediation in an- titrust violations. While the Department of Justice may pursue criminal and civil penalties for firms and their employees for engaging in illegal anti-competitive behavior, secondary civil suits may result in large payouts for victims.6 Lande and Davis (2011) estimate a value of $6.8 to $7.7 billion for federal antitrust remedies from 1990 through 2007. They compare this to recoveries from 40 of the largest private antitrust cases over the same time period, which total $18.0 to $19.6 billion.7 They conclude that private enforcement exhibits a stronger deterrent effect on antitrust violations due to the size of recoveries.
6Civil anti-trust laws also resemble False Claims Act statutes; both include treble damages and may be brought by private third parties. States also frequently have their own anti-trust statutes, in addition to the federal Sherman and Clayton Acts. 7These figures come from an earlier study of private anti-trust actions. See Lande and Davis (2007).
66 Tax evasion and anti-trust violations differ from health care fraud in that they represent a transfer from the government to the evader or from victims to firms. Health care fraud committed against the government not only acts as a transfer, but also has potential impacts on health through inefficient levels of health care provision to patients.
2.2.2 The Social Optimality of Whistleblower Laws
The social optimality of whistleblower litigation and private “bounties” has been widely discussed.8 Whistleblower laws incorporate private individuals seeking profit through the judicial system, and while this “outsourcing” of fraud monitoring from government agencies to private individuals can dramatically increase the resources utilized in anti-fraud activities, it also inevitably introduces additional distortions. Whistleblowers may be over-zealous in reporting potential fraud cases, some of which will be dropped and others which may be settled out of court to avoid uncertainy from a jury’s verdict. There is mixed evidence that whistleblowers overcrowd the Attorney General with poor cases. Matthew (2007) argues that pharmaceutical manufacturers may be the target of a flood of unmeritorious cases initiated by FCA whistleblowers, and cites the large number of suits against pharmaceutical manufacturers and staggering settlements as evidence. Kwok (2013) analyzes whistleblower- initiated FCA cases in search of evidence of high-volume, low effort “filing mill” strategies pursued by plaintiff attorneys, but finds no indication that this behavior is common.
2.2.3 Empirical Studies on Health Care Fraud in Public Programs
There has been limited work which attempts to empirically analyze the economic effects of anti-fraud enforcement activity in public programs. DOJ officials only provide anecdotal evidence of FCA cases preventing fraud. Total “Health Care Fraud and Abuse Control” program investments resulted in a return-on-investment of $4.90 to $1.00 in fiscal years
2006 to 2008, and $7.90 to $1.00 in fiscal years 2010 and 2012 (United States Government
Accountability Office, 2013). These estimates exclude any deterrent effect. Similarly, in a
8See Breit and Elzinga (1985) as private enforcement relates to antitrust actions and Depoorter and De Mot (2006), Broderick (2007), Matthew (2007), Heyes and Kapur (2009), Engstrom (2013), and Kwok (2013) related to the False Claims Act.
67 survey of prosecutors in state Attorney General’s offices in states with and without their own FCAs, most respondents did not know if the statute is effective in deterring fraud
(Bucy et al., 2010).
One case study, Kesselheim, Darby, et al. (2011), attempts to quantify the deterrent effect of FCA enforcement on prescription rates in a lawsuit against Warner-Lambert, a pharmaceutical manufacturing company and producer of Neurontin. They find that upward trends in the number of new prescriptions and spending for Gabapentin continued during all phases of the DOJ investigation and these trends were greatest for off-label uses. Only after settlement of the case did growth rates become negative, but this occurred for both on and off-label uses.
Becker, Kessler, and McClellan (2005) measure the impact of anti-fraud activity on abusive behavior by examining Medicare patient spending and health outcomes for groups of illnesses identified as being vulnerable to fraud and abuse. State anti-fraud activity is measured as state MFCU expenditures per Medicare beneficiary, and alternatively as
MFCU expenditures per hospital. The authors find no overall aggregate effect of state anti- fraud activity, but certain patient groups and providers were found to be more responsive, with a greater response to anti-fraud activity in patient populations for whom additional treatment and expenditures would be of marginal benefit.
The aggregate result of no effect of Medicaid anti-fraud activity on Medicare patients raises a question about overlapping program administration and enforcement. States have little motivation to focus MFCU expenditures on Medicare patients, as Medicare is federally funded, while Medicaid spending is jointly financed by the state and federal government.
2.3 The False Claims Act
Various studies summarize the federal and state FCAs, which we highlight here.9 State
statutes generally mirror the federal FCA with subtle differences. The original 1863 statute
contained a “qui tam,” or whistleblower, provision which allowed private parties to bring
9Barger et al. (2005), Depoorter and De Mot (2006), Broderick (2007), Doyle (2009), Bucy et al. (2010), and Kwok (2013), in addition to others.
68 a suit, even if they, themselves, were not directly affected by the fraudulent perpetration.
The FCA was infrequently used until 1986 when amendments strengthened the qui tam provisions and made it easier to file a suit as well as substantially increased the civil damages for a violation.
2.3.1 FCA Utilization and Recoveries
Statistics from the U.S. Department of Justice show the growing importance of the FCA, especially in health care fraud suits (DOJ, 2016). Figures 2.1 and 2.2 show the number of
“new matters” filed since 1986 and associated settlements and judgment amounts. New matters include new referrals, investigations and qui tam actions. In 1987, the number of cases filed in which the Department of Health and Human Services (HHS) was the primary client included only about 4 percent of total cases. In 2015, HHS cases represent roughly
60 percent of total cases and 55 percent of settlements and judgments. Since 2000, HHS cases have made up about 56 percent of all FCA cases and 72 percent of total settlements and judgments.10 Similarly, the number of qui tam actions has grown from only a few per year to representing the majority of cases filed, growing from 20 percent of HHS cases in
1987 to 94 percent in 2015. Since 2010, total settlements, judgments, and awards averaged over $4 billion annually. Whistleblower-initiated FCA case filings surpass private antitrust
filings (Sourcebook of Criminal Justice Statistics, 2012) and settlements and judgment dollar amounts rival that of private securities class action suits (Bulan, Ryan, and Simmons,
2016).11
2.3.2 Recent Amendments
Since the 1986 amendments of the FCA, the federal government has heavily relied on the FCA to combat fraud, and this is further evidenced in more recent amendments of the statute. The 2005 Deficit Reduction Act provides states with an incentive to enact their
10This is true of FCA suits filed by states as well. In a survey of state FCAs, most states reported that they primarily use their FCA to pursue health care cases (Bucy et al., 2010). 11On average, from 2010 through 2012 there were 559 private civil anti-trust cases filed in district courts annually. Settlements from class action securities lawsuits from 2010 - 2015 average $3 billion annually.
69 own FCAs which follow the federal statute, by receiving an additional 10 percent of any recoveries (HHS, ). Between 2005 and 2012, 12 states enacted a FCA and as of 2012, 12 states’ FCAs were compliant with the Deficit Reduction Act (NAMFCU, 2006/2015). In
2009, the Fraud Enforcement and Recovery Act further strengthened the 1986 amendments of the FCA, and in 2010 the Patient Protection and Affordable Care Act changed the federal
FCA by allowing a violation of the federal anti-kickback statute to automatically trigger a violation of the False Claims Act; until the ACA amendment, it was unclear whether a violation of the anti-kickback statute should result in a per se violation of the FCA (Becker,
2014).
2.3.3 FCA Characteristics, Filing Procedure, and the Role of the Whistle-
blower
In a typical FCA suit, a sealed complaint submitted by a qui tam plaintiff is reviewed by the DOJ and a United States attorney, who consider the case’s merit and evidence.
A qui tam plaintiff need not be directly injured by the violation. The DOJ, or state
Attorney General, must then decide whether to intervene in the case or not. In the case of government intervention, the qui tam party may remain a plaintiff in the suit, but the government assumes primary responsibility in prosecuting the case. The qui tam plaintiff may still conduct the action even if the DOJ or state Attorney General chooses not to intervene. Generally, a qui tam plaintiff is awarded anywhere from 15 to 25 percent of the total damages if the government intervenes and 25 to 30 percent if the government does not.12 Although relator awards are slightly smaller in cases where the government intervenes, private plaintiffs gain the significant resources of the state or federal government, and success rates indicate that intervention and resulting settlements and judgments are positively correlated; 95 percent of FCA qui tam cases between 1986 and 2009 where the
DOJ intervened resulted in settlements and judgments, compared to only 9 percent of non-intervened cases (Kwok, 2013). Dismissal rates of cases where only a qui tam relator
12Only under specific scenarios is a qui tam plaintiff awarded less, for example if the relator provides publicly disclosed information as evidence or if he or she is partially at fault for the violation.
70 proceeds are much higher, about 80 percent, versus a dismissal rate of 4 percent when the government is also a plaintiff in the suit (Matthew, 2007). Importantly, damages for violations of the FCA include civil monetary penalties of $5,500 to $11,000 per violation plus treble damages. The damage awards may quickly grow to multi-million dollar suits which can potentially attract private plaintiffs who wish to gain up to one-third of the total damage amount.
2.4 Conceptual Issues: Why Prescription Drugs?
2.4.1 FCAs and Different Types of Fraud
It is likely that state FCAs and other anti-fraud activities affect various beneficiary groups and service types differently. We focus on prescription drugs in our analysis.13 In particular, we focus on off-label pharmaceutical marketing because characteristics of both
FCAs and prescription drug markets make pharmaceutical manufacturers a lucrative target for prosecution. Profit motives, whistleblower rewards, and per-violation monetary penalties and treble damages combined with high-volume transactions and deep-pocketed defendants in the pharmaceutical industry all contribute to the FCAs disproportionately large effect on prescription drugs.
Off-label Marketing
Off-label marketing occurs when a pharmaceutical manufacturer engages in marketing efforts towards providers or patients that promote a drug for uses or populations other than what has been approved by the FDA. In a typical off-label marketing case, FCA damages are based upon the inducement of health care providers to prescribe a drug for uses or populations which it was not tested and approved for (Gaier, Scher, and Sharma,
13Pharmaceuticals are not the only service that is prone to fraud and abuse. Home health care and durable medical equipment are two other service areas that are highlighted (HHS, 2012). In home health care fraud, providers may falsely verify the necessity of home health services, receive kickbacks for signing beneficiaries, or verify medical necessity for patients not under the physician’s care. Other common fraudulent behavior includes overstating the severity of patient’s conditions and billing for services that were not provided (HHS and DOJ, 2016). Home health care violations constitute only about 5.5 percent of civil settlements and judgments, and make up only a small percentage of FCA lawsuits (OIG, 2000/2015; Qureshi et al., 2011; Kesselheim and Studdert, 2008).
71 2013). This type of illegal behavior falls under a FCA violation because the government will only authorize payments for drugs which are safe, effective, and if it will be used for a
“medically accepted indication,” such as a use approved by the FDA. Otherwise, if a drug is marketed for some other unapproved use, then it arguably does not satisfy this criteria and the government ends up paying for falsely advertised products (Samuelson, 2014). Off- label marketing violations are prosecutable by states under FCAs and/or by using MFCUs
(Spacapan and Hutchison, 2013; OIG, 2000/2015).
Consequences of off-label marketing include increasing health care costs, possibly en- dangering public health, and undermining federal drug regulation. Off-label marketing may provide benefit for some populations. Various anti-depressants, for example, are prescribed for a number of unapproved uses including anxiety, obsessive-compulsive disorders, eating disorders, and PTSD, among other uses (Wittich, Burkle, and Lanier, 2012). Off-label pre- scribing provides a faster alternative path for effective treatment when medical knowledge outpaces the FDA approval process and gives physicians flexibility when a standard treat- ment might fail (Klein and Tabarrok, 2004). Note that physicians may prescribe a drug for any reason; FCA prosecution rarely targets physicians for their (off-label) prescribing behavior, but rather is used to limit pharmaceutical manufacturers’ influence on physician prescription behavior through promotion of unproven drug benefits or uses (Samuelson,
2014).
High Transaction Volume
One common characteristic of the FCA is that it provides for a civil penalty per violation of the statute, in addition to treble damages. This per-violation penalty would matter more for service categories where the number of transactions is relatively high. One such type of case is off-label marketing cases. In an off-label marketing case, each prescription that is written based upon illegal marketing constitutes a transaction, and consequently civil penalties may quickly add up.14 Damages resulting from off-label marketing or over-
14Under an “implied certification” liability, all prescriptions after the submission of a false claim (or incidence of illegal off-label marketing) will be treated as “false claims,” not just those prescriptions induced by the off-label marketing.
72 charging for thousands of prescriptions will result in higher payouts than a typical home health provider, for example, who is double billing for services. Examining MFCU cases in
2014, 48 home health care agency and 413 home health aide criminal convictions resulted in $7.1 million and $12.5 million in recoveries, which is an average of about $150,000 and
$30,000 per case. This is paltry compared to the nearly $2.8 million average recovery for a civil case involving a pharmaceutical manufacturer.
Deep-Pocketed Defendants
Cases involving pharmaceutical manufacturers may also entail large settlements or judg- ments due to the deep pockets of such companies and wide reach of any violations. For ex- ample, Johnson & Johnson, in 2013, was ordered to pay over $2.2 billion in civil and criminal penalties for off-label promotion of Risperdal, in addition to other drugs, as well as kickbacks to health care providers (DOJ, 2013). Office of the Inspector General (OIG) reports show that pharmaceutical manufacturers and pharmacies are the most common provider type in civil MFCU cases, representing approximately two-thirds of settlements and judgments and recoveries (OIG, 2000/2015). Kesselheim and Studdert (2008) find that although phar- maceutical manufacturers represent only 4 percent of defendants in whistleblower-initiated
FCA cases between 1996 and 2005, this group constitutes almost 40 percent of recoveries.
The authors note that the number of cases where hospitals or physician groups are defen- dants declined over the study period; this may be due to strategic enforcement by the DOJ where prosecuting hospitals, which made up 29 percent of the sample of cases but only
3 percent of recoveries, is forgone in favor of more lucrative cases against pharmaceutical manufacturers.
Insiders and Detection
Off-label promotion by pharmaceutical manufacturers may also engage more potential whistleblowers in the illegal activity and increase the probability of detection. Kesselheim,
Mello, and Studdert (2011) find that in 18 federal FCA fraud cases, 71 percent of whistle- blowers worked as pharmaceutical sales representatives and another 20 percent worked as sales or accounting managers. Kesselheim and Studdert (2008) find that about 77 percent of
73 whistleblowers in FCA health care fraud cases are “insiders.” If off-label marketing activity involves more individuals, including sales representatives, managers, and executives, than a physician office where only the physician, patient, and possibly a billing specialist could report false billing, then it may be more difficult for the firm to collude with its employees and discourage whistleblowing.
FCAs provide a clear profit motive for whistleblowers to target cases in which expected awards will be the greatest. Evidence from MFCU cases suggests that damages in pharma- ceutical cases are far larger, on average, than damages in other cases. Therefore, there is reason to suspect that not all providers and service types should respond similarly to state anti-fraud activity, and further, if civil penalties and enforcement should have a measurable effect on behavior, we should expect to see responses in the prescription drug service area.
2.4.2 Policy Confounders
We hypothesize that state Medicaid prescription drug spending should be more respon- sive to FCA legislation, relative to other spending categories. Focusing on prescription drug spending, it is necessary to consider other state Medicaid policies which may influence spending. If these state-level policies are correlated with FCA implementation, then omit- ting them will bias the estimated effect of state FCAs on prescription drug spending and we will attribute reductions in prescription drug spending to FCAs rather than the other state policies.
Prescription drug spending is a common target for state Medicaid cost-containment policies. Although provision of prescription drug benefits is optional, all fifty states and the
District of Columbia choose to provide these services. As such, states have greater flexibility in designing their prescription drug benefits, relative to mandatory services (Schneider and
Elam, 2002). Periods of rapid prescription drug spending growth have also gained the attention of states looking to control Medicaid program costs. Medicaid drug expenditures grew at 19 percent annually between 1997 and 2000 (Schneider and Elam, 2002). From
1999 to 2009, the average annual number of prescriptions per non-dual beneficiary remained stable, yet the average annual costs per beneficiary nearly doubled (Verdier and Zlatinov,
2013). After a decade of stable growth, and a slight dip in Medicaid prescription drug
74 expenditures, prescription drug spending growth increased sharply in 2014. This most recent increase in drug expenditures is attributable to high-cost specialty drugs entering the market and ACA Medicaid expansions (MACPAC, 2016; Martin et al., 2016).
States have implemented numerous policies to curb program spending on prescription drugs. Common policies include preferred drug lists (PDLs), prior authorization (PA), cost-sharing with Medicaid enrollees, prescription limits, drug category exclusions, generic substitution laws, and supplemental manufacturer rebates (Sourmerai, 2004; Morden and
Sullivan, 2005). Generally, PDLs, PA, cost-sharing, and generic substitution seek to in-
fluence the choice of prescription drugs and shift utilization towards more cost effective products. Prescription limits and category exclusions directly limit prescription drug cov- erage.
PDLs enumerate drugs that state agencies indicate as being the most cost-effective or efficient. Drugs which are not listed on a state’s PDL are still available to beneficiaries, but prescribers must go through a prior authorization process and request approval for each prescription. States may also exercise PDLs as a negotiating tool with pharmaceutical manufacturers to demand rebates that exceed the required federal rebates (Mellow, Stud- dert, and Brennan, 2004). Co-payments represent another policy which a state may use to reduce overall utilization or influence beneficiary choice. Medicaid co-payments are limited to nominal amounts and a beneficiary may not be denied a prescription due to his or her inability to pay. However, states may implement tiered co-payments, where generic drugs typically have a lower co-payment than branded drugs, in order to attempt and shift utiliza- tion toward less costly generic drugs (Hoadley, 2005). Generic substitution laws, while not specific to Medicaid programs, also encourage the use of low-cost generics, when available.
Laws vary by state, but typically allow a pharmacist to dispense a generic drug, even if a physician has written a prescription for the brand name (Shrank et al., 2010).
Medicaid programs may exclude certain broad categories of drugs including drugs for weight loss, infertility, cosmetic and hair growth, smoking cessation, vitamins, and cold remedies among others (Hoadley, 2005). States may individually choose any or all of these categories for which to exclude Medicaid coverage. Prescription limits are limits on either the quantity of filled prescriptions (i.e. 30 days or 45 days) or on the total amount of
75 prescriptions per beneficiary per month. Prescription limits more commonly apply to all prescriptions, but as of 2010, 4 states had implemented “brand-only” prescription limits in an attempt to also shift utilization to generic drugs (Lieberman et al., 2016).
States have adopted various forms of policies to control Medicaid prescription drug costs.
We account for these common prescription drug policies in our empirical specifications.
Specific state program variables and data sources are discussed in sections 5 and 6.
2.4.3 State and Federal FCA Overlap
The presence of a federal FCA, in addition to secondary state-specific FCAs begs the question of the overlap of roles of both statutes in anti-fraud activity and why a state FCA, which is no more punitive than the federal FCA, would additionally deter fraud. We argue that state FCAs represent additional resources devoted by the state to detect and prosecute fraud and provide greater opportunities for whistleblowers to disclose potential fraud. State
FCAs (most of which contain whistleblower provisions) provide additional channels through which whistleblowers may come forward and provide information on potential fraudulent behavior. Instead of submitting a case directly to the DOJ, whistleblowers may file com- plaints to their own state (typically the Attorney General) who then may go forward with the case under the state FCA or submit to the DOJ to involve the federal government in the prosecution. Many states with FCAs also report substantial cooperation with the state
MFCU (Bucy et al., 2010). Additional resources and channels for whistleblowers to disclose potential fraud increase the probability of detection and prosecution at the state level.
Our analysis focuses on prescription drugs and we hypothesize that state FCAs should deter illegal off-label marketing and/or prescribing. Why should off-label marketing by pharmaceutical manufacturers respond to state anti-fraud efforts? Although pharmaceuti- cal marketing campaigns (direct-to-consumer advertising) are not tailored for each state, pharmaceutical manufacturers do target advertising by consumer demographics.15 These
types of promotions may potentially exhibit a response to states who target allegedly de-
15See Alpert, Lakdawalla, and Sood (2015) who uses variation in direct-to-consumer advertising after Medicare part D implementation and across markets with high versus low elderly population shares to identify the effect of advertising on drug utilization.
76 ceptive advertising more aggressively. Illegal off-label promotion, however, occurs primar- ily through physician detailing by pharmaceutical sales representatives and is physician- specific. Detailing practices may respond differentially, depending on the level of anti-fraud activity within a state. Direct-to-physician detailing makes up a significant proportion of pharmaceutical company marketing budgets, and this form of promotion has been found to influence physician prescription behavior (Datta and Dave, 2017). Although we do not focus on lawsuits dealing with pharmaceutical companies’ misrepresentations of prescrip- tion drug costs to state Medicaid programs to secure higher reimbursements, this type of fraud is also liable under the FCA. The presence of a state FCA places a higher cost on inaccurate reporting of drug prices or generic status to the state Medicaid program.
2.5 Data and Descriptive Statistics
Our analysis combines four data sources containing information on FCA legislation,
MFCU expenditures, Medicaid spending and eligible populations, and demographics. State
FCAs and MFCU expenditures vary at the state-year level, and we combine these datasets with Medicaid spending measures by group and service category within each state and year.
The resulting data set contains state anti-fraud activity and Medicaid spending from 1999 through 2012 for all 50 states and D.C. State Medicaid prescription drug policies are also included in specifications to control for potentially confounding policy effects.
Data on state FCA statutes was collected by analyzing state laws and amendments over time. Information on state statutes was primarily gathered using Lexis Advance legal database. Although there are several documents which catalog various state FCAs, none compiles all of the characteristics of the laws or how the laws have been amended over time.16
A detailed summary of state FCA statutes is provided in the appendix. The statutes in each state were individually summarized and standardized to facilitate comparisons across states.
For each state with a FCA enacted, common characteristics of the law are documented as well as amendments over time. Characteristics of state FCAs include the year that the
16See Barger et al. (2005), Rosenbaum, Lopez, and Stifler (2009), Bucy et al. (2010), and NAMFCU (2006/2015).
77 statute was enacted, whether the statute covers all government programs or just health care (Medicaid), whether the statute has a qui tam, or “whistleblower”, provision, the reward for a whistleblower plaintiff, and the monetary penalty for violating the statute. See
Table 2.1 for a list of states which have enacted a FCA and the year of enactment; Figures
2.3 and 2.4 summarize FCA implementation over time. States which have enacted FCAs tend to be dispersed geographically, and states have gradually implemented FCAs over time.
Between 1991 and 2016, typically one to two FCAs were enacted each year. The years 2006 to 2007 saw the most activity with 5 statutes being passed. Just over half of FCAs passed by
1999 applied to state Medicaid program fraud only, but the majority of FCAs passed since apply to all state government programs. Similarly, most FCAs passed since 1991 contain a whistleblower provision, with more than 80 percent having a whistleblower component in
2016, compared to 60 percent in 1991 (Figure 2.4).
Following Becker, Kessler, and McClellan (2005), we also include a measure of state
MFCU expenditures to additionally control for state-level anti-fraud activity. State MFCU expenditures come from “Office of the Inspector General Medicaid Fraud Control Unit An- nual Reports” for years 1999 through 2003 and 2009 through 2012. Unfortunately, the
OIG MFCU annual reports change format beginning in 2004 and only aggregate program expenditures are included. MFCU expenditures for 2004 and 2005 are based upon reported aggregate program grants for state programs and estimated average state shares of the total program grants. The National Association of Medicaid Fraud Control Units (NAMFCU) began conducting a survey of state MFCU programs in 2006. This survey includes annual budgeted amounts for state MFCU programs for years 2006 through 2015. We use this sec- ondary source to fill in missing years from the OIG reports by adjusting down the budgeted amount to fit trends in the OIG-reported data. Figure 2.6 shows how the two MFCU ex- penditure data sources compare, along with our imputed measures of MFCU spending. We use OIG spending amounts for every year available, 2004 and 2005 are imputed from state shares of the total program budget, and predicted values are used for years 2006 through
2009. State MFCU expenditures are adjusted using the consumer price index (CPI) and are expressed in 2012 dollars.
Medicaid spending data is from the Medicaid Statistical Information System (MSIS).
78 MSIS state reporting became compulsory in 1999. States report Medicaid claims to CMS who then produce utilization and program statistics for that state. The MSIS data spans from fiscal year (beginning October 1) 1999 through 2012 and includes total spending by group, or “basis of eligibility,” which includes children, adults, disabled/blind, aged, and a total category, and service type which includes 31 categories of services such as inpatient hospital, prescription drugs, and physician services. One category, religious services, which is only present in one year of the data is excluded. All spending figures are adjusted to 2012 dollars using the Medical Care CPI index.
Combining state, year, eligibility group, and service categories, our dataset has 107,100 observations (51 states and D.C. × 14 years × 5 groups × 30 categories). However, we choose to focus on a few select service categories in this paper: “total,” “total excluding prescription drugs,” and “prescription drugs only.” Total and prescription drug categories are included in the MSIS data, and we construct an additional category that excludes prescription drugs by taking the difference of spending in the the total and prescription drug categories. We calculate eligibles at the state-year level as the total number of eligibles for each group, and
“all except aged” is calculated as the sum of children, adult, and disabled eligibles.17
MSIS data are available for most states in all years.18 Federal Medical Assistance
Percentages (FMAPs) are obtained from the Federal Register, state unemployment data comes from the BLS Local Area Unemployment Statistics, and population estimates are from the U.S. Census Bureau Intercensal Estimates.
State prescription drug policy variables are constructed using the National Pharmaceu- tical Council’s (NPC) “Pharmaceutical Benefits under State Medical Assistance Programs” annual reports from 1999 through 2007. The NPC annual reports are based upon a survey
17The sum of eligible counts and spending amounts across individual groups does not equal the eligible count or spending amount “total” category because groups such as “unknown” and eligibles through the Breast and Cervical Cancer Prevention Act are not included in the data. On average, our groups represent 90 percent plus of eligibles and spending. Additionally, eligible counts reflect unique individual counts where an individual is counted separately in each service category if they received that type of service, but only once in the “total” category. 18Maine is excluded in 2011, and in 2012 Arizona, Colorado, District of Columbia, Florida, Hawaii, Idaho, Kansas, Louisiana, Massachusetts, Maine, Texas, and Utah are excluded due to data availability. To balance the panel, we assume 2011 values for states missing data in 2012.
79 of states and include state-by-state tables of current policies.19 We supplement the NPC data with Medicaid prescription drug benefits data from the Kaiser Family Foundation and Medicaid Analytic Extract statistical compendiums. Kaiser benefits data are avail- able for years 2003, 2004, 2006, 2008, 2010, and 2012, and the Medicaid Analytic Extract compendiums are available from 1999 through 2009.
Table 2.2 summarizes expenditure variables and other model covariates over time. Dur- ing the sample window, 22 states enacted their own FCAs, increasing the number of states with a FCA statute from 14 in 1999 to 36 states in 2012. As expected, average total state
Medicaid expenditures increase over time, yet spending per eligible decreases. MFCU ex- penditures increase in total, but remain relatively stable on a per Medicaid eligible basis.
Table 2.3 summarizes variables in 2012, separately for states which enacted their own FCA, both ever and limiting to enactment only during the sample period 1999 through 2012, and for states which have never enacted their own FCA.
States which have enacted a FCA have, on average, larger Medicaid programs ($8.3 billion versus $4.8 billion), but spending per eligible is slightly lower ($5,285 versus $5,854).
States that enact a FCA after 1999 have, on average, slightly higher Medicaid spending per eligible ($5,420).20 FCA states also have larger MFCU expenditures, both in total (about
$4.1 million to $5.3 million compared to $1.8 million) and per Medicaid eligible ($3.32 to
$3.39 versus $2.53). This is consistent with states who pass a FCA having a stronger stance on Medicaid fraud control and may indicate that FCA legislation and MFCU spending are complements in state fraud control efforts. States with FCAs have more eligibles enrolled in managed care, have a higher percentage of black population and slightly more Democrats in the state Senate and House.
Notably, states with a FCA enacted typically have a lower FMAP on average (56.2 to 58.1 versus 63.3). This could reflect relatively richer states, which consequently have lower FMAPs, enacting FCAs and using a greater pool of resources to combat fraud. A lower FMAP also means that the state is paying relatively more, dollar for dollar, for its
19Simon, Tennyson, and Hudman (2009) perform a similar analysis on aggregate Medicaid prescription drug spending and state Medicaid policies using NPC annual reports. 20Medicaid spending measures reflect spending for all service types and groups combined.
80 Medicaid program and receives less matching from the federal government. Consequently, states with lower FMAPs have a stronger financial incentive to target fraud in their own
Medicaid programs. This financial incentive is absent for Medicare. Figure 2.5 explores the relationship between state FMAPs and FCA implementation further. By grouping states into quartiles based on their relative FMAPs in 1991, we see that states with a lower
FMAP, which implies less generous matching by the federal government, tend to lead other states in passing FCAs. After about 1999, states with FMAPs below the 25th percentile begin to rapidly enact FCA legislation, and states with FMAPs between the 25th and 50th percentile behave similarly. States with the highest FMAPs and which receive the most generous federal matching for their Medicaid programs lag behind other states in passing
FCA legislation, with only 42 percent of states with an FMAP above the 75th percentile passing a FCA by 2016, compared to 92 percent of states in the first quartile and 77 percent in the second quartile of FMAPs.21 These trends lend some support to the idea that states view anti-fraud enforcement as a money-saving endeavor, through direct recoveries from fraud settlements and judgments and through the deterrence of future fraud.
Table 2.4 shows changes in state prescription drug policies over time. Policies include prior authorization (PA), whether or not the state’s Medicaid program requires a co-pay for prescriptions, pharmacy reimbursement formula, prescription limits (as categorized by
Lieberman et al. (2016)), and the generic substitution rate, which is the proportion of pre- scriptions filled with a generic drug in cases where a generic is available. According to Table
2.4, states are increasingly implementing prior authorization policies, requiring co-pays, es- tablishing brand-specific prescription limits and overall prescription limits, and substituting more generics. States are also adopting more complex pharmacy reimbursement formulas which incorporate wholesale acquisition cost (WAC), actual acquisition cost (AAC), or a combination of these and average wholesale price (AWP). By 2012, only 35 percent of states have a pharmacy reimbursement policy that uses AWP as the only benchmark, compared to 88 percent in 1999. This is consistent with evidence from a federal audit in 2000 that
21Figure 2.5 looks very similar if instead of calculating FMAP percentiles based on 1991 FMAPs, we use a state’s average FMAP from 1991 through 2016.
81 found AWP benchmark prices to be inflated and thus not a useful benchmark for Medicaid reimbursement (Alpert, Duggan, and Hellerstein, 2013).
Figures 2.7 a and 2.8 present total Medicaid spending and Medicaid spending on pre- scription drugs, by eligible group.22 Generally, the disabled and aged groups make up the two largest spending groups. Total spending for all groups, except children, dips after 2005, and then continues to grow at a slightly slower rate. The dip in spending is most dramatic for the aged and disabled groups and is due in large part to the shift of dually eligible in- dividuals to Medicare as Medicare Part D was implemented. Consistent with this program change, spending on prescription drugs exhibits a more dramatic drop post-2005.23 Over the whole sample period, total prescription drug spending averages about $35 billion per year. Spending hit a peak in 2005 at $55 billion and was about $24 billion in 2012. Figure
2.9 shows how the total number of aged eligibles differentially responds during this period.
While the number of eligibles in other groups continues to increase or dips slightly, the num- ber of aged eligibles noticeably drops before returning to a similar rate of growth. Given the Medicare program changes which differentially affected aged eligibles and spending, we suspect that any potential effect of anti-fraud activities on Medicaid spending on aged ben- eficiaries may be confounded by the Medicare Modernization Act. Although spending for aged and disabled significantly drops, spending for adults and children remains relatively stable, only decreasing slightly after 2011, when certain ACA provisions were implemented which extended drug rebate offers to Medicaid managed care plans and as states shifted drug spending to managed care plans (MACPAC, 2016).
Table 2.5 presents mean statistics for total spending, Medicaid eligibles, and spending per eligible over all states and years 1999 through 2012 by service type and group. For all service categories shown, the disabled represent the single largest group in terms of total spending and spending per eligible. Children represent the largest group in Medicaid eligibles. Spending per eligible for “all services,” on average, ranges from just over $2,000
22Medicaid prescription drug spending data from the MSIS does not include rebates paid under the federal drug rebate program. 23Bruen and Miller, 2008 estimate that Medicaid spending on prescription drugs fell by about 50 percent on average after Medicare Part-D was implemented.
82 for children to over $16,700 for disabled. Prescription drug spending per eligible ranges from $155 for children to over $2,300 for disabled.
2.6 Empirical Model
2.6.1 Difference-in-Difference Model Specification
We estimate a generalized difference-in-differences model shown in equation (1).
0 log(yst) = fcastγ + log(mfcust)φ + xstβ + ηs + ξt + εst (2.1)
In this specification, the log of Medicaid spending per eligible, yst, for state s in year t, is regressed on an indicator variable for whether the state has enacted a FCA as of year t,
the log of state MFCU expenditures per Medicaid eligible, state demographic and Medicaid
program controls, and state and year fixed effects. Models are stratified by eligible groups.
The coefficients of interest are γ and φ on the FCA indicator and (logged) MFCU expen-
diture per beneficiary. The vector x includes the state’s FMAP in year t, unemployment
rate, log of gross state product, population controls such as the proportion of the total
state population under age 14, over age 65, female age 15 to 44, and proportion black, state
welfare program characteristics including the percent of uninsured low income children and
minimum wage and Medicaid program controls including age distribution controls, man-
aged care enrollment, and non-dual eligibles interacted with a Medicare part-D dummy (0
prior to 2006 and 1 for 2006 onward), and state government controls including proportion
of the state senate and house that is Democrat. Following Howard (2010), state-level demo-
graphics are included to control for factors that may influence Medicaid program spending.
State welfare and government controls proxy for program generosity and state attitudes
towards welfare which likely affect Medicaid spending as well. In models with prescription
drug spending as the dependent variable, we also include state prescription drug policies.
Policies include indicators for prior authorization, whether or not the state’s Medicaid pro-
gram requires a co-pay for prescriptions, “AWP only” pharmacy reimbursement formula,
prescription limits, and the generic substitution rate. Standard errors are clustered at the
83 state level in all specifications (Bertrand, Duflo, and Mullainathan, 2004).
One may be concerned with selection of states choosing to enact their own FCA. This choice may be due to unobservable characteristics, and it may be inappropriate to compare states which have a FCA statute to states which do not. We can consider two potential scenarios. First, states with the highest spending, and potentially the most fraudulent activity, will enact their own FCA in order to control spending. States with their own
FCAs would then be associated with higher spending per beneficiary, as compared to states with no such statute, and the coefficient estimate on the parameters of interest in the naive model would be biased toward zero or positive.
Second, states which enact their own FCA likely choose to do so because they benefit
financially from its passage. Although states collect larger payouts if their FCA is harmo- nized with the federal FCA statute, they may decide to conserve administrative resources and free-ride on federal or other state anti-fraud efforts, Bucy et al. (2010). According to a 2014 OIG report, two-thirds of civil cases prosecuted with MFCU resources, which frequently utilize FCAs, are “global” cases where multiple states are plaintiffs in the suit.
This likely induces a type of free-rider behavior where a single state need not enact its own
FCA or heavily invest in anti-fraud activity to become plaintiff in a suit and benefit from a group-effort to prosecute defrauders. As a result of free-riding behavior, the deterrent effect of FCAs will be underestimated. We correct for this unobserved heterogeneity by us- ing state-specific fixed effects (Besley and Case, 2010; Galiani, Gertler, and Schargrodsky,
2005). This method will control for any time-invariant characteristics that may influence a state’s decision to enact an FCA.
2.6.2 Aggregate Medicaid Expenditure Results
Table 2.6 shows specifications for categories “All services,” “Excluding Prescription
Drugs,” and “Prescription Drugs Only.” All regressions include year and state fixed ef- fects, as well as group compositional controls in models where multiple eligibility groups are included. In each table, the first column presents results for all Medicaid eligibility groups (except aged). The next columns examine the effect of state anti-fraud activity and legislation on spending for children, adults, and the disabled separately, then all groups
84 combined (including aged), and finally aged alone. The first column omits the aged group because this group may exhibit differential Medicaid program participation trends due to passage of the Medicare Modernization Act.24
Panel 1 of Table 2.6 (all services) shows that generally, there is no effect of state FCAs or MFCU expenditures on aggregate Medicaid spending per eligible.25 There is a positive and significant (10 percent level) of FCA legislation on spending per adult eligible, and a negative 7 percent effect for the disabled.26 Panel 2 presents results for the “excluding prescription drugs” category. Results for “excluding prescription drugs” are similar to those of all services combined. There is a significant (10 percent level) negative effect of FCA legislation for the disabled and aged groups, and positive effect for adults.
Specifications in Panel 3 include prescription drug spending only. Results indicate much larger (negative) effects of state FCAs. For all groups, controlling for composition effects, there is a 18.3 percent decline in prescription drug spending per eligible (significant at the 5 percent level). This effect is slightly larger for the “all except aged” category, which implies a 20.6 percent decline in prescription drug spending per eligible (significant at the 5 percent level). There is also a significant 17.6 percent decline in spending for the disabled. Across all specifications, the sign and magnitude of the estimated effect of state FCAs is similar, with the aged specification showing the smallest effect and is insignificant. State MFCU expenditures have no significant effect on spending in any specification although the point estimates are negative for all groups except adult and aged. This may reflect the imputation of MFCU dollars in some years.
Panel 4 further includes state prescription drug policy variables. FCA coefficients are
24Although we attempt and control for the implementation of Medicare Part D, the magnitude of this policy still likely introduces noise into our model. We should only be concerned with comparing across pre- and post-Medicare Part D years if states which have implemented a FCA exhibit systematically different responses in prescription drug spending than those which have not. Trends in spending for FCA versus non-FCA states do not seem to suggest this type of bias. Limiting the sample to years 2006 and forward does reduce magnitudes slightly. Coefficient estimates on the FCA term maintain a negative sign, but standard errors inflate since we are throwing out approximately half of the identifying variation in state FCA legislation. 25See appendix Tables B.1 and B.2 for specifications including additional services and sample periods. 26All estimated coefficients except the log of MFCU expenditures have been transformed by (eβ −1)×100 and may be interpreted as a percentage change. The log of MFCU expenditures is interpreted as an elasticity with regard to a 10 percent increase in MFCU expenditures.
85 extremely similar. Medicaid co-payments have a large and negative effect, but only in the
“Child” category. “AWP-only” reimbursement is associated with increased prescription drug spending per eligible. This is consistent with findings from Alpert, Duggan, and Hellerstein
(2013) that AWP benchmarks were inflated by manufacturers, and as states moved away from AWP reimbursement benchmarks, drug reimbursements fell.
Results in panels 3 and 4 of Table 2.6 are markedly different than those for all services aggregated and excluding prescription drugs. Although standard errors are relatively large, the responsiveness of prescription drug spending per eligible to state FCAs is much greater than in other categories shown. These results are consistent with incentives contained within state FCA legislation and provide evidence that the presence of state anti-fraud legislation has a substantial negative effect on prescription drug spending. Applying results from the
“all (except aged)” specification, state FCA legislation reduces prescription drug spending by $115 per eligible (21.2 percent X $543), on average, resulting in an average annual savings to a state of $107 million. In total, this implies about $5.4 billion in annual savings on Med- icaid prescription drug expenditures if all states were to implement an FCA. This savings to the Medicaid program reflects a behavioral response of pharmaceutical manufacturers, pharmacies, and providers and is a savings in addition to any actual recoveries, settlements, and judgments from FCA or other fraud prosecutions. Total federal FCA recoveries where the Department of Health and Human Services was a plaintiff average about $2.6 billion annually between 2010 and 2016, and OIG reports show that at the state level, between one to two-thirds of civil recoveries related to Medicaid Fraud Control Unit activity are as- sociated with prosecution of pharmaceutical manufacturers. The deterrent effect estimated in this paper is large in comparison to actual recoveries in health care fraud cases.
The estimated effects of state FCAs on prescription drug spending are sizable. Estimates of health care fraud suggest that up to 10 percent of total expenditures are fraudulent, im- plying almost $300 billion in fraud and abuse in 2015. Medicaid and Medicare are identified as being vulnerable populations for fraud and abuse, and fraud enforcement activity sug- gests that fraudulent activity is not evenly distributed across spending categories (HHS,
2012; Rosenbaum, Lopez, and Stifler, 2009). If MFCU cases target pharmaceutical man- ufacturers in almost two-thirds of cases, then presumably there is a heightened level of
86 perceived fraud being perpetrated by pharmaceutical companies. Although off-label pre- scriptions are not themselves illegal, rates of off-label prescription use averaged 21 percent of all prescriptions in 2001, but for some classes of medicines ranged from 46 percent to 81 percent of prescriptions (Radley, Finkelstein, and Stafford, 2009). In summary, we expect the effects on prescription drug spend to be large.
2.7 Off-Label Usage: State Drug Utilization Data
Our analysis of aggregate state Medicaid spending suggests that state FCAs reduce total prescription drug spending with little effect on other spending categories. We use State
Drug Utilization Data (SDUD) from CMS to test whether there is evidence of a behavioral response in prescribing patterns. Alleged off-label marketing forms the basis of FCA liability for pharmaceutical manufacturers in many cases (Qureshi et al., 2011; Kesselheim, Mello, and Studdert, 2011). If FCAs do deter off-label marketing and subsequent prescribing, then we should expect to find a relatively larger effect for drugs which are prone to off-label use versus those that are not.
2.7.1 SDUD
State Drug Utilization Data from CMS includes prescription drug expenditures and uti- lization for state Medicaid programs beginning in 1991.27 States report utilization data on covered outpatient prescription drugs in compliance with the federal Medicaid Drug Rebate program. Spending and utilization data are available by state, quarter, and National Drug
Code (NDC) which identifies a unique prescription drug product down to the package size.
Expenditures in the SDUD are gross of any rebates paid by pharmaceutical manufactur- ers to state Medicaid programs, and current releases of the data censor state-quarter-NDC observations with 10 or less prescriptions.
We aggregate quarterly observations into annual totals by state and annualize in cases where one or more quarters of data are missing. From 1999 through 2012 there are al-
27SDUD has been utilized in numerous previous studies including Duggan and Scott Morton (2006); Kesselheim, Fischer, and Avorn (2006); Fischer, Choudhry, and Winkelmayer (2007); Bruen and Miller (2008); Alpert, Duggan, and Hellerstein (2013); and Bradford and Bradford (2017) among others.
87 most 5 million state-year-NDC observations. We exclude certain quarters of data where total expenditures and prescriptions are implausibly high or volatile, similar to Bradford and Bradford (2017). Following Alpert, Duggan, and Hellerstein (2013), we omit outliers by comparing spending per prescription for a given NDC.28 We match each NDC into a therapeutic class (primary and subclass) using Medispan’s Generic Product Identifier hi- erarchical classification system, and aggregate to the state-year-subclass level. Table 2.7 presents therapeutic classes ranked by total prescription volume in the State Drug Uti- lization Data (SDUD) from 1999 through 2012. Antidepressants make up the largest single category, representing just over 6 percent of prescriptions in the dataset. The top five classes, antidepressants, opioid analgesics, antiasthmatics, anticonvulsants, and antihypertensives make up just over a quarter of total prescriptions, and three-quarters of all prescriptions are represented by the top 24 classes. In some model specifications, we limit the sample to the top 40 therapeutic classes, which make up over 90 percent of prescription volume.29
2.7.2 Therapeutic Class and Off-Label Use
To see if there is evidence of a behavioral response of pharmaceutical manufacturers or prescribers to FCA prosecution of off-label promotion, we categorize therapeutic classes as
“off-label prone” and “not off-label prone.” We base these classifications on previous litera- ture within the medical field which examines off-label use and drug product characteristics.
In an investigative analysis of the top 15 therapeutic classes in terms of spending, choles- terol medications were noted as being rarely prescribed for unapproved treatments while
28We omit Arkansas 1991q1 and 2004q4, Colorado 2000q2 and 2001q2, Idaho 2005q1, Indiana 2000q1, 2000q3, and 2012q1, Iowa 1991q1 through 2001q2, Kansas 1999q4 and 2000q1, Kentucky 2012q1 through 2012q4, Maine 2008q1, Maryland 1999q1, 2000q2, 2000q3, 2000q4, Minnesota 2000q1, Missouri 2007q4, Nevada 2002q4, North Dakota 2005q2, Ohio 2010q1 through 2011q3, Pennsylvania 2004q2, Rhode Island 2006q4 and 2007q4 through 2012q4, South Dakota 1991q1 through 2002q3, and 2007q3, Vermont 1999q2 through 1999q4, and 2004q2, and Washington 2003q4, 2006q3, and 2009q3. We exclude Arizona and Ten- nessee from the sample completely. Data for Arizona is unavailable for many years, since Arizona does not participate in the drug rebate program, and many quarters for TN appear to have poor data quality with large and obvious erroneous outliers. We also trim observations above the 99 percentile of spending per prescription within an NDC across all states and years and limit the sample of spending and prescriptions to Medicaid fee-for-service (FFS) observations omit managed care (MCO) spending and prescription obser- vations which are available only beginning in 2010. Our sample restrictions leave about 96 percent of the FFS state-quarter-NDC observations in the SDUD. 29Ranking classes by total spending, rather than prescriptions, results in a very similar ordering of drug classes.
88 three-quarters of anti-seizure medications, two-thirds of antipsychotics, and one-quarter of antidepressants are prescribed for off-label uses (Adams and Young, 2003). Radley, Finkel- stein, and Stafford (2009) analyze physician-patient encounter data and find that off-label use varies widely by therapeutic class. Controlling for other drug-specific characteristics, cardiac therapies, including antianginals, antiarrhythmics, and anticoagulants, are found to be 6.8 times more likely to be prescribed for an off-label indication, relative to analgesics.
Other therapeutic classes associated with higher off-label use include anticonvulsants, psy- chiatric therapies including antidepressants, anxiolytics, and antipsychotics, allergy thera- pies, antiasthmatics, and ulcer and dyspepsia medications. Off-label usage rates for these classes of drugs range from 30 to 46 percent. Diabetes therapies were found to be associ- ated with very little off-label use with only 1 percent of drug mentions for off-label uses.
Off-label use of analgesics, agents to lower lipid levels, hormone therapies, contraception, and antihypertensives were also low, ranging from 6 to 14 percent of drug mentions.
Walton et al. (2008) find evidence of substantial off-label prescribing, despite little evi- dence of efficacy, for drugs within the therapeutic classes of antidepressants, antipsychotics, and anxiolytic-sedatives. Chen et al. (2005) specifically examine off-label use of anticon- vulsants among Medicaid beneficiaries and find that over 70 percent of patients taking an anticonvulsant received a prescription for an off label indication. Lin, Phan, and Lin
(2006) document off-label usage of beta-blockers and estimate that on average 52 percent of prescriptions are for off-label uses.
Table 2.8 lists our categorization of therapeutic classes as “off-label prone” versus “not off-label prone” based on reviewed literature. Although only about a quarter of observations fall into the off-label prone or not off-label prone categories, these categories represent about half of total spending and prescriptions. Appendix Table 3A provides extensive detail on how we categorize therapeutic classes into off-label categories. In addition to our review of medical and pharmaceutical literature on off-label use, we incorporate estimated off-label
89 rates provided by Bradford, Paker, and Williams (2015).30 We include in our analysis additional model specifications which use a continuous probability of off-label use, which varies by therapeutic class and subclass. Figure 2.10 shows the distribution of average estimated off-label rates. Very few classes have an extremely low or high estimated rate of off-label use; 50 percent of total prescriptions fall within therapeutic classes (and subclasses) with an an average estimated off-label rate between about 30 and 50 percent. The average off-label rate across all drug classes is 43.8 percent. For classes which the medical literature indicates are prone to off-label use, the average estimated off-label rate is 44.0 percent with the Bradford, Paker, and Williams estimates. For classes indicated as not prone the average rate is 22.9 percent.
2.7.3 Model
We construct a triple-difference model which estimates the effect of state FCAs for implementing versus non-implementing states, before and after implementation, for off- label prone classes of drugs versus not off-label prone. Our specification generally follows
Hoynes, Schanzenbach, and Almond (2016) who have a similar generalized difference-in- difference model that is extended to a triple-difference model with an additional margin of variation.